Add Bitnet model (#37742)

* Adding BitNet b1.58 Model

* Add testing code for BitNet

* Fix format issues

* Fix docstring format issues

* Fix docstring

* Fix docstring

* Fix: weight back to uint8

* Fix

* Fix format issues

* Remove copy comments

* Add model link to the docstring

* Fix: set tie_word_embeddings default to false

* Update

* Generate modeling file

* Change config name for automatically generating modeling file.

* Generate modeling file

* Fix class name

* Change testing branch

* Remove unused param

* Fix config docstring

* Add docstring for BitNetQuantConfig.

* Fix docstring

* Update docs/source/en/model_doc/bitnet.md

Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com>

* Update docs/source/en/model_doc/bitnet.md

Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>

* Update bitnet config

* Update explanation between online and offline mode

* Remove space

* revert changes

* more revert

* spaces

* update

* fix-copies

* doc fix

* fix minor nits

* empty

* small nit

* empty

---------

Co-authored-by: Shuming Ma <shumingma@pku.edu.cn>
Co-authored-by: shumingma <shmingm@gmail.com>
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com>

.circleci/config.yml

@@ -154,7 +154,7 @@ jobs:
                   path: ~/transformers/installed.txt
             - run: python -c "from transformers import *" || (echo '🚨 import failed, this means you introduced unprotected imports! 🚨'; exit 1)
             - run: ruff check examples tests src utils
-            - run: ruff format tests src utils --check
+            - run: ruff format examples tests src utils --check
             - run: python utils/custom_init_isort.py --check_only
             - run: python utils/sort_auto_mappings.py --check_only
             - run: python utils/check_doc_toc.py

.circleci/create_circleci_config.py

@@ -30,9 +30,28 @@ COMMON_ENV_VARIABLES = {
     "RUN_PIPELINE_TESTS": False,
 }
 # Disable the use of {"s": None} as the output is way too long, causing the navigation on CircleCI impractical
-COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "dist": "loadfile", "vvv": None, "rsfE":None}
+COMMON_PYTEST_OPTIONS = {"max-worker-restart": 0, "vvv": None, "rsfE":None}
 DEFAULT_DOCKER_IMAGE = [{"image": "cimg/python:3.8.12"}]
 
+# Strings that commonly appear in the output of flaky tests when they fail. These are used with `pytest-rerunfailures`
+# to rerun the tests that match these patterns.
+FLAKY_TEST_FAILURE_PATTERNS = [
+    "OSError",  # Machine/connection transient error
+    "Timeout",  # Machine/connection transient error
+    "ConnectionError",  # Connection transient error
+    "FileNotFoundError",  # Raised by `datasets` on Hub failures
+    "PIL.UnidentifiedImageError",  # Raised by `PIL.Image.open` on connection issues
+    "HTTPError",  # Also catches HfHubHTTPError
+    "AssertionError: Tensor-likes are not close!",  # `torch.testing.assert_close`, we might have unlucky random values
+    # TODO: error downloading tokenizer's `merged.txt` from hub can cause all the exceptions below. Throw and handle
+    # them under a single message.
+    "TypeError: expected str, bytes or os.PathLike object, not NoneType",
+    "TypeError: stat: path should be string, bytes, os.PathLike or integer, not NoneType",
+    "Converting from Tiktoken failed",
+    "KeyError: <class ",
+    "TypeError: not a string",
+]
+
 
 class EmptyJob:
     job_name = "empty"
@@ -124,7 +143,9 @@ class CircleCIJob:
                 # Examples special case: we need to download NLTK files in advance to avoid cuncurrency issues
         timeout_cmd = f"timeout {self.command_timeout} " if self.command_timeout else ""
         marker_cmd = f"-m '{self.marker}'" if self.marker is not None else ""
-        additional_flags = f" -p no:warning -o junit_family=xunit1 --junitxml=test-results/junit.xml"
+        junit_flags = f" -p no:warning -o junit_family=xunit1 --junitxml=test-results/junit.xml"
+        joined_flaky_patterns = "|".join(FLAKY_TEST_FAILURE_PATTERNS)
+        repeat_on_failure_flags = f"--reruns 5 --reruns-delay 2 --only-rerun '({joined_flaky_patterns})'"
         parallel = f' << pipeline.parameters.{self.job_name}_parallelism >> '
         steps = [
             "checkout",
@@ -150,9 +171,10 @@ class CircleCIJob:
                     "command": f"TESTS=$(circleci tests split  --split-by=timings {self.job_name}_test_list.txt) && echo $TESTS > splitted_tests.txt && echo $TESTS | tr ' ' '\n'" if self.parallelism else f"awk '{{printf \"%s \", $0}}' {self.job_name}_test_list.txt > splitted_tests.txt"
                     }
             },
+            {"run": {"name": "fetch hub objects before pytest", "command": "python3 utils/fetch_hub_objects_for_ci.py"}},
             {"run": {
                 "name": "Run tests",
-                "command": f"({timeout_cmd} python3 -m pytest {marker_cmd} -n {self.pytest_num_workers} {additional_flags} {' '.join(pytest_flags)} $(cat splitted_tests.txt) | tee tests_output.txt)"}
+                "command": f"({timeout_cmd} python3 -m pytest {marker_cmd} -n {self.pytest_num_workers} {junit_flags} {repeat_on_failure_flags} {' '.join(pytest_flags)} $(cat splitted_tests.txt) | tee tests_output.txt)"}
             },
             {"run": {"name": "Expand to show skipped tests", "when": "always", "command": f"python3 .circleci/parse_test_outputs.py --file tests_output.txt --skip"}},
             {"run": {"name": "Failed tests: show reasons",   "when": "always", "command": f"python3 .circleci/parse_test_outputs.py --file tests_output.txt --fail"}},
@@ -185,6 +207,9 @@ torch_job = CircleCIJob(
 generate_job = CircleCIJob(
     "generate",
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
+    # networkx==3.3 (after #36957) cause some issues
+    # TODO: remove this once it works directly
+    install_steps=["uv venv && uv pip install . && uv pip install networkx==3.2.1"],
     marker="generate",
     parallelism=6,
 )
@@ -248,6 +273,7 @@ examples_torch_job = CircleCIJob(
     docker_image=[{"image":"huggingface/transformers-examples-torch"}],
     # TODO @ArthurZucker remove this once docker is easier to build
     install_steps=["uv venv && uv pip install . && uv pip install -r examples/pytorch/_tests_requirements.txt"],
+    pytest_num_workers=4,
 )
 
 
@@ -255,6 +281,7 @@ examples_tensorflow_job = CircleCIJob(
     "examples_tensorflow",
     additional_env={"OMP_NUM_THREADS": 8},
     docker_image=[{"image":"huggingface/transformers-examples-tf"}],
+    pytest_num_workers=2,
 )
 
 
@@ -305,6 +332,9 @@ repo_utils_job = CircleCIJob(
 non_model_job = CircleCIJob(
     "non_model",
     docker_image=[{"image": "huggingface/transformers-torch-light"}],
+    # networkx==3.3 (after #36957) cause some issues
+    # TODO: remove this once it works directly
+    install_steps=["uv venv && uv pip install . && uv pip install networkx==3.2.1"],
     marker="not generate",
     parallelism=6,
 )
@@ -334,9 +364,9 @@ doc_test_job = CircleCIJob(
     pytest_num_workers=1,
 )
 
-REGULAR_TESTS = [torch_job, tf_job, flax_job, hub_job, onnx_job, tokenization_job, processor_job, generate_job, non_model_job] # fmt: skip
-EXAMPLES_TESTS = [examples_torch_job, examples_tensorflow_job]
-PIPELINE_TESTS = [pipelines_torch_job, pipelines_tf_job]
+REGULAR_TESTS = [torch_job, flax_job, hub_job, onnx_job, tokenization_job, processor_job, generate_job, non_model_job] # fmt: skip
+EXAMPLES_TESTS = [examples_torch_job]
+PIPELINE_TESTS = [pipelines_torch_job]
 REPO_UTIL_TESTS = [repo_utils_job]
 DOC_TESTS = [doc_test_job]
 ALL_TESTS = REGULAR_TESTS + EXAMPLES_TESTS + PIPELINE_TESTS + REPO_UTIL_TESTS + DOC_TESTS + [custom_tokenizers_job] + [exotic_models_job]  # fmt: skip

.github/ISSUE_TEMPLATE/bug-report.yml

@@ -38,24 +38,30 @@ body:
 
           - text models: @ArthurZucker
           - vision models: @amyeroberts, @qubvel
-          - speech models: @ylacombe, @eustlb
+          - speech models: @eustlb
           - graph models: @clefourrier
 
         Library:
 
-          - flax: @sanchit-gandhi
+          - flax: @gante and @Rocketknight1
           - generate: @zucchini-nlp (visual-language models) or @gante (all others)
           - pipelines: @Rocketknight1
           - tensorflow: @gante and @Rocketknight1
           - tokenizers: @ArthurZucker and @itazap
-          - trainer: @muellerzr @SunMarc
+          - trainer: @zach-huggingface @SunMarc
 
         Integrations:
 
-          - deepspeed: HF Trainer/Accelerate: @muellerzr
+          - deepspeed: HF Trainer/Accelerate: @SunMarc @zach-huggingface
           - ray/raytune: @richardliaw, @amogkam
           - Big Model Inference: @SunMarc
           - quantization (bitsandbytes, autogpt): @SunMarc @MekkCyber
+        
+        Devices/Backends:
+        
+          - AMD ROCm: @ivarflakstad
+          - Intel XPU: @IlyasMoutawwakil
+          - Ascend NPU: @ivarflakstad 
 
         Documentation: @stevhliu
 
@@ -72,7 +78,7 @@ body:
 
         Maintained examples (not research project or legacy):
 
-          - Flax: @sanchit-gandhi
+          - Flax: @Rocketknight1
           - PyTorch: See Models above and tag the person corresponding to the modality of the example.
           - TensorFlow: @Rocketknight1
 

.github/ISSUE_TEMPLATE/i18n.md

@@ -23,7 +23,7 @@ Some notes:
 * Please translate in a gender-neutral way.
 * Add your translations to the folder called `<languageCode>` inside the [source folder](https://github.com/huggingface/transformers/tree/main/docs/source).
 * Register your translation in `<languageCode>/_toctree.yml`; please follow the order of the [English version](https://github.com/huggingface/transformers/blob/main/docs/source/en/_toctree.yml).
-* Once you're finished, open a pull request and tag this issue by including #issue-number in the description, where issue-number is the number of this issue. Please ping @stevhliu and @MKhalusova for review.
+* Once you're finished, open a pull request and tag this issue by including #issue-number in the description, where issue-number is the number of this issue. Please ping @stevhliu for review.
 * 🙋 If you'd like others to help you with the translation, you can also post in the 🤗 [forums](https://discuss.huggingface.co/).
 
 ## Get Started section

.github/PULL_REQUEST_TEMPLATE.md

@@ -41,22 +41,22 @@ Models:
 
 - text models: @ArthurZucker
 - vision models: @amyeroberts, @qubvel
-- speech models: @ylacombe, @eustlb
+- speech models: @eustlb
 - graph models: @clefourrier
 
 Library:
 
-- flax: @sanchit-gandhi
+- flax: @gante and @Rocketknight1
 - generate: @zucchini-nlp (visual-language models) or @gante (all others)
 - pipelines: @Rocketknight1
 - tensorflow: @gante and @Rocketknight1
 - tokenizers: @ArthurZucker
-- trainer: @muellerzr and @SunMarc
+- trainer: @zach-huggingface and @SunMarc
 - chat templates: @Rocketknight1
 
 Integrations:
 
-- deepspeed: HF Trainer/Accelerate: @muellerzr
+- deepspeed: HF Trainer/Accelerate: @SunMarc @zach-huggingface
 - ray/raytune: @richardliaw, @amogkam
 - Big Model Inference: @SunMarc
 - quantization (bitsandbytes, autogpt): @SunMarc @MekkCyber
@@ -72,7 +72,7 @@ HF projects:
 
 Maintained examples (not research project or legacy):
 
-- Flax: @sanchit-gandhi
+- Flax: @Rocketknight1
 - PyTorch: See Models above and tag the person corresponding to the modality of the example.
 - TensorFlow: @Rocketknight1
 

.github/scripts/assign_reviewers.py

@@ -0,0 +1,120 @@
+# coding=utf-8
+# Copyright 2025 the HuggingFace Inc. team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import os
+import github
+import json
+from github import Github
+import re
+from collections import Counter
+from pathlib import Path
+
+def pattern_to_regex(pattern):
+    if pattern.startswith("/"):
+        start_anchor = True
+        pattern = re.escape(pattern[1:])
+    else:
+        start_anchor = False
+        pattern = re.escape(pattern)
+    # Replace `*` with "any number of non-slash characters"
+    pattern = pattern.replace(r"\*", "[^/]*")
+    if start_anchor:
+        pattern = r"^\/?" + pattern  # Allow an optional leading slash after the start of the string
+    return pattern
+
+def get_file_owners(file_path, codeowners_lines):
+    # Process lines in reverse (last matching pattern takes precedence)
+    for line in reversed(codeowners_lines):
+        # Skip comments and empty lines, strip inline comments
+        line = line.split('#')[0].strip()
+        if not line:
+            continue
+
+        # Split into pattern and owners
+        parts = line.split()
+        pattern = parts[0]
+        # Can be empty, e.g. for dummy files with explicitly no owner!
+        owners = [owner.removeprefix("@") for owner in parts[1:]]
+
+        # Check if file matches pattern
+        file_regex = pattern_to_regex(pattern)
+        if re.search(file_regex, file_path) is not None:
+            return owners  # Remember, can still be empty!
+    return []  # Should never happen, but just in case
+
+def pr_author_is_in_hf(pr_author, codeowners_lines):
+    # Check if the PR author is in the codeowners file
+    for line in codeowners_lines:
+        line = line.split('#')[0].strip()
+        if not line:
+            continue
+
+        # Split into pattern and owners
+        parts = line.split()
+        owners = [owner.removeprefix("@") for owner in parts[1:]]
+
+        if pr_author in owners:
+            return True
+    return False
+
+def main():
+    script_dir = Path(__file__).parent.absolute()
+    with open(script_dir / "codeowners_for_review_action") as f:
+        codeowners_lines = f.readlines()
+
+    g = Github(os.environ['GITHUB_TOKEN'])
+    repo = g.get_repo("huggingface/transformers")
+    with open(os.environ['GITHUB_EVENT_PATH']) as f:
+        event = json.load(f)
+
+    # The PR number is available in the event payload
+    pr_number = event['pull_request']['number']
+    pr = repo.get_pull(pr_number)
+    pr_author = pr.user.login
+    if pr_author_is_in_hf(pr_author, codeowners_lines):
+        print(f"PR author {pr_author} is in codeowners, skipping review request.")
+        return
+
+    existing_reviews = list(pr.get_reviews())
+    if existing_reviews:
+        print(f"Already has reviews: {[r.user.login for r in existing_reviews]}")
+        return
+
+    users_requested, teams_requested = pr.get_review_requests()
+    users_requested = list(users_requested)
+    if users_requested:
+        print(f"Reviewers already requested: {users_requested}")
+        return
+
+    locs_per_owner = Counter()
+    for file in pr.get_files():
+        owners = get_file_owners(file.filename, codeowners_lines)
+        for owner in owners:
+            locs_per_owner[owner] += file.changes
+
+    # Assign the top 2 based on locs changed as reviewers, but skip the owner if present
+    locs_per_owner.pop(pr_author, None)
+    top_owners = locs_per_owner.most_common(2)
+    print("Top owners", top_owners)
+    top_owners = [owner[0] for owner in top_owners]
+    try:
+        pr.create_review_request(top_owners)
+    except github.GithubException as e:
+        print(f"Failed to request review for {top_owners}: {e}")
+
+
+
+if __name__ == "__main__":
+    main()

.github/scripts/codeowners_for_review_action

@@ -0,0 +1,370 @@
+# Top-level rules are matched only if nothing else matches
+* @Rocketknight1 @ArthurZucker # if no one is pinged based on the other rules, he will do the dispatch
+*.md @stevhliu
+*tokenization* @ArthurZucker
+docs/ @stevhliu
+/benchmark/ @McPatate
+/docker/ @ydshieh @ArthurZucker
+
+# More high-level globs catch cases when specific rules later don't apply
+/src/transformers/models/*/processing* @molbap @yonigozlan @qubvel
+/src/transformers/models/*/image_processing* @qubvel
+/src/transformers/models/*/image_processing_*_fast* @yonigozlan
+
+# Owners of subsections of the library
+/src/transformers/generation/ @gante
+/src/transformers/pipeline/ @Rocketknight1 @yonigozlan
+/src/transformers/integrations/ @SunMarc @MekkCyber @zach-huggingface
+/src/transformers/quantizers/ @SunMarc @MekkCyber
+tests/ @ydshieh
+tests/generation/ @gante
+
+/src/transformers/models/auto/ @ArthurZucker
+/src/transformers/utils/ @ArthurZucker @Rocketknight1
+/src/transformers/loss/ @ArthurZucker
+/src/transformers/onnx/ @michaelbenayoun
+
+# Specific files come after the sections/globs, so they take priority
+/.circleci/config.yml @ArthurZucker @ydshieh
+/utils/tests_fetcher.py @ydshieh
+trainer.py @zach-huggingface @SunMarc
+trainer_utils.py @zach-huggingface @SunMarc
+/utils/modular_model_converter.py @Cyrilvallez @ArthurZucker
+
+# Owners of individual models are specific / high priority, and so they come last
+# mod* captures modeling and modular files
+
+# Text models
+/src/transformers/models/albert/mod*_albert* @ArthurZucker
+/src/transformers/models/bamba/mod*_bamba* @ArthurZucker
+/src/transformers/models/bart/mod*_bart* @ArthurZucker
+/src/transformers/models/barthez/mod*_barthez* @ArthurZucker
+/src/transformers/models/bartpho/mod*_bartpho* @ArthurZucker
+/src/transformers/models/bert/mod*_bert* @ArthurZucker
+/src/transformers/models/bert_generation/mod*_bert_generation* @ArthurZucker
+/src/transformers/models/bert_japanese/mod*_bert_japanese* @ArthurZucker
+/src/transformers/models/bertweet/mod*_bertweet* @ArthurZucker
+/src/transformers/models/big_bird/mod*_big_bird* @ArthurZucker
+/src/transformers/models/bigbird_pegasus/mod*_bigbird_pegasus* @ArthurZucker
+/src/transformers/models/biogpt/mod*_biogpt* @ArthurZucker
+/src/transformers/models/blenderbot/mod*_blenderbot* @ArthurZucker
+/src/transformers/models/blenderbot_small/mod*_blenderbot_small* @ArthurZucker
+/src/transformers/models/bloom/mod*_bloom* @ArthurZucker
+/src/transformers/models/bort/mod*_bort* @ArthurZucker
+/src/transformers/models/byt5/mod*_byt5* @ArthurZucker
+/src/transformers/models/camembert/mod*_camembert* @ArthurZucker
+/src/transformers/models/canine/mod*_canine* @ArthurZucker
+/src/transformers/models/codegen/mod*_codegen* @ArthurZucker
+/src/transformers/models/code_llama/mod*_code_llama* @ArthurZucker
+/src/transformers/models/cohere/mod*_cohere* @ArthurZucker
+/src/transformers/models/cohere2/mod*_cohere2* @ArthurZucker
+/src/transformers/models/convbert/mod*_convbert* @ArthurZucker
+/src/transformers/models/cpm/mod*_cpm* @ArthurZucker
+/src/transformers/models/cpmant/mod*_cpmant* @ArthurZucker
+/src/transformers/models/ctrl/mod*_ctrl* @ArthurZucker
+/src/transformers/models/dbrx/mod*_dbrx* @ArthurZucker
+/src/transformers/models/deberta/mod*_deberta* @ArthurZucker
+/src/transformers/models/deberta_v2/mod*_deberta_v2* @ArthurZucker
+/src/transformers/models/dialogpt/mod*_dialogpt* @ArthurZucker
+/src/transformers/models/diffllama/mod*_diffllama* @ArthurZucker
+/src/transformers/models/distilbert/mod*_distilbert* @ArthurZucker
+/src/transformers/models/dpr/mod*_dpr* @ArthurZucker
+/src/transformers/models/electra/mod*_electra* @ArthurZucker
+/src/transformers/models/encoder_decoder/mod*_encoder_decoder* @ArthurZucker
+/src/transformers/models/ernie/mod*_ernie* @ArthurZucker
+/src/transformers/models/ernie_m/mod*_ernie_m* @ArthurZucker
+/src/transformers/models/esm/mod*_esm* @ArthurZucker
+/src/transformers/models/falcon/mod*_falcon* @ArthurZucker
+/src/transformers/models/falcon3/mod*_falcon3* @ArthurZucker
+/src/transformers/models/falcon_mamba/mod*_falcon_mamba* @ArthurZucker
+/src/transformers/models/fastspeech2_conformer/mod*_fastspeech2_conformer* @ArthurZucker
+/src/transformers/models/flan_t5/mod*_flan_t5* @ArthurZucker
+/src/transformers/models/flan_ul2/mod*_flan_ul2* @ArthurZucker
+/src/transformers/models/flaubert/mod*_flaubert* @ArthurZucker
+/src/transformers/models/fnet/mod*_fnet* @ArthurZucker
+/src/transformers/models/fsmt/mod*_fsmt* @ArthurZucker
+/src/transformers/models/funnel/mod*_funnel* @ArthurZucker
+/src/transformers/models/fuyu/mod*_fuyu* @ArthurZucker
+/src/transformers/models/gemma/mod*_gemma* @ArthurZucker
+/src/transformers/models/gemma2/mod*_gemma2* @ArthurZucker
+/src/transformers/models/glm/mod*_glm* @ArthurZucker
+/src/transformers/models/openai_gpt/mod*_openai_gpt* @ArthurZucker
+/src/transformers/models/gpt_neo/mod*_gpt_neo* @ArthurZucker
+/src/transformers/models/gpt_neox/mod*_gpt_neox* @ArthurZucker
+/src/transformers/models/gpt_neox_japanese/mod*_gpt_neox_japanese* @ArthurZucker
+/src/transformers/models/gptj/mod*_gptj* @ArthurZucker
+/src/transformers/models/gpt2/mod*_gpt2* @ArthurZucker
+/src/transformers/models/gpt_bigcode/mod*_gpt_bigcode* @ArthurZucker
+/src/transformers/models/gptsan_japanese/mod*_gptsan_japanese* @ArthurZucker
+/src/transformers/models/gpt_sw3/mod*_gpt_sw3* @ArthurZucker
+/src/transformers/models/granite/mod*_granite* @ArthurZucker
+/src/transformers/models/granitemoe/mod*_granitemoe* @ArthurZucker
+/src/transformers/models/herbert/mod*_herbert* @ArthurZucker
+/src/transformers/models/ibert/mod*_ibert* @ArthurZucker
+/src/transformers/models/jamba/mod*_jamba* @ArthurZucker
+/src/transformers/models/jetmoe/mod*_jetmoe* @ArthurZucker
+/src/transformers/models/jukebox/mod*_jukebox* @ArthurZucker
+/src/transformers/models/led/mod*_led* @ArthurZucker
+/src/transformers/models/llama/mod*_llama* @ArthurZucker @Cyrilvallez
+/src/transformers/models/longformer/mod*_longformer* @ArthurZucker
+/src/transformers/models/longt5/mod*_longt5* @ArthurZucker
+/src/transformers/models/luke/mod*_luke* @ArthurZucker
+/src/transformers/models/m2m_100/mod*_m2m_100* @ArthurZucker
+/src/transformers/models/madlad_400/mod*_madlad_400* @ArthurZucker
+/src/transformers/models/mamba/mod*_mamba* @ArthurZucker
+/src/transformers/models/mamba2/mod*_mamba2* @ArthurZucker
+/src/transformers/models/marian/mod*_marian* @ArthurZucker
+/src/transformers/models/markuplm/mod*_markuplm* @ArthurZucker
+/src/transformers/models/mbart/mod*_mbart* @ArthurZucker
+/src/transformers/models/mega/mod*_mega* @ArthurZucker
+/src/transformers/models/megatron_bert/mod*_megatron_bert* @ArthurZucker
+/src/transformers/models/megatron_gpt2/mod*_megatron_gpt2* @ArthurZucker
+/src/transformers/models/mistral/mod*_mistral* @ArthurZucker
+/src/transformers/models/mixtral/mod*_mixtral* @ArthurZucker
+/src/transformers/models/mluke/mod*_mluke* @ArthurZucker
+/src/transformers/models/mobilebert/mod*_mobilebert* @ArthurZucker
+/src/transformers/models/modernbert/mod*_modernbert* @ArthurZucker
+/src/transformers/models/mpnet/mod*_mpnet* @ArthurZucker
+/src/transformers/models/mpt/mod*_mpt* @ArthurZucker
+/src/transformers/models/mra/mod*_mra* @ArthurZucker
+/src/transformers/models/mt5/mod*_mt5* @ArthurZucker
+/src/transformers/models/mvp/mod*_mvp* @ArthurZucker
+/src/transformers/models/myt5/mod*_myt5* @ArthurZucker
+/src/transformers/models/nemotron/mod*_nemotron* @ArthurZucker
+/src/transformers/models/nezha/mod*_nezha* @ArthurZucker
+/src/transformers/models/nllb/mod*_nllb* @ArthurZucker
+/src/transformers/models/nllb_moe/mod*_nllb_moe* @ArthurZucker
+/src/transformers/models/nystromformer/mod*_nystromformer* @ArthurZucker
+/src/transformers/models/olmo/mod*_olmo* @ArthurZucker
+/src/transformers/models/olmo2/mod*_olmo2* @ArthurZucker
+/src/transformers/models/olmoe/mod*_olmoe* @ArthurZucker
+/src/transformers/models/open_llama/mod*_open_llama* @ArthurZucker
+/src/transformers/models/opt/mod*_opt* @ArthurZucker
+/src/transformers/models/pegasus/mod*_pegasus* @ArthurZucker
+/src/transformers/models/pegasus_x/mod*_pegasus_x* @ArthurZucker
+/src/transformers/models/persimmon/mod*_persimmon* @ArthurZucker
+/src/transformers/models/phi/mod*_phi* @ArthurZucker
+/src/transformers/models/phi3/mod*_phi3* @ArthurZucker
+/src/transformers/models/phimoe/mod*_phimoe* @ArthurZucker
+/src/transformers/models/phobert/mod*_phobert* @ArthurZucker
+/src/transformers/models/plbart/mod*_plbart* @ArthurZucker
+/src/transformers/models/prophetnet/mod*_prophetnet* @ArthurZucker
+/src/transformers/models/qdqbert/mod*_qdqbert* @ArthurZucker
+/src/transformers/models/qwen2/mod*_qwen2* @ArthurZucker
+/src/transformers/models/qwen2_moe/mod*_qwen2_moe* @ArthurZucker
+/src/transformers/models/rag/mod*_rag* @ArthurZucker
+/src/transformers/models/realm/mod*_realm* @ArthurZucker
+/src/transformers/models/recurrent_gemma/mod*_recurrent_gemma* @ArthurZucker
+/src/transformers/models/reformer/mod*_reformer* @ArthurZucker
+/src/transformers/models/rembert/mod*_rembert* @ArthurZucker
+/src/transformers/models/retribert/mod*_retribert* @ArthurZucker
+/src/transformers/models/roberta/mod*_roberta* @ArthurZucker
+/src/transformers/models/roberta_prelayernorm/mod*_roberta_prelayernorm* @ArthurZucker
+/src/transformers/models/roc_bert/mod*_roc_bert* @ArthurZucker
+/src/transformers/models/roformer/mod*_roformer* @ArthurZucker
+/src/transformers/models/rwkv/mod*_rwkv* @ArthurZucker
+/src/transformers/models/splinter/mod*_splinter* @ArthurZucker
+/src/transformers/models/squeezebert/mod*_squeezebert* @ArthurZucker
+/src/transformers/models/stablelm/mod*_stablelm* @ArthurZucker
+/src/transformers/models/starcoder2/mod*_starcoder2* @ArthurZucker
+/src/transformers/models/switch_transformers/mod*_switch_transformers* @ArthurZucker
+/src/transformers/models/t5/mod*_t5* @ArthurZucker
+/src/transformers/models/t5v1.1/mod*_t5v1.1* @ArthurZucker
+/src/transformers/models/tapex/mod*_tapex* @ArthurZucker
+/src/transformers/models/transfo_xl/mod*_transfo_xl* @ArthurZucker
+/src/transformers/models/ul2/mod*_ul2* @ArthurZucker
+/src/transformers/models/umt5/mod*_umt5* @ArthurZucker
+/src/transformers/models/xmod/mod*_xmod* @ArthurZucker
+/src/transformers/models/xglm/mod*_xglm* @ArthurZucker
+/src/transformers/models/xlm/mod*_xlm* @ArthurZucker
+/src/transformers/models/xlm_prophetnet/mod*_xlm_prophetnet* @ArthurZucker
+/src/transformers/models/xlm_roberta/mod*_xlm_roberta* @ArthurZucker
+/src/transformers/models/xlm_roberta_xl/mod*_xlm_roberta_xl* @ArthurZucker
+/src/transformers/models/xlm_v/mod*_xlm_v* @ArthurZucker
+/src/transformers/models/xlnet/mod*_xlnet* @ArthurZucker
+/src/transformers/models/yoso/mod*_yoso* @ArthurZucker
+/src/transformers/models/zamba/mod*_zamba* @ArthurZucker
+
+# Vision models
+/src/transformers/models/beit/mod*_beit* @amyeroberts @qubvel
+/src/transformers/models/bit/mod*_bit* @amyeroberts @qubvel
+/src/transformers/models/conditional_detr/mod*_conditional_detr* @amyeroberts @qubvel
+/src/transformers/models/convnext/mod*_convnext* @amyeroberts @qubvel
+/src/transformers/models/convnextv2/mod*_convnextv2* @amyeroberts @qubvel
+/src/transformers/models/cvt/mod*_cvt* @amyeroberts @qubvel
+/src/transformers/models/deformable_detr/mod*_deformable_detr* @amyeroberts @qubvel
+/src/transformers/models/deit/mod*_deit* @amyeroberts @qubvel
+/src/transformers/models/depth_anything/mod*_depth_anything* @amyeroberts @qubvel
+/src/transformers/models/depth_anything_v2/mod*_depth_anything_v2* @amyeroberts @qubvel
+/src/transformers/models/deta/mod*_deta* @amyeroberts @qubvel
+/src/transformers/models/detr/mod*_detr* @amyeroberts @qubvel
+/src/transformers/models/dinat/mod*_dinat* @amyeroberts @qubvel
+/src/transformers/models/dinov2/mod*_dinov2* @amyeroberts @qubvel
+/src/transformers/models/dinov2_with_registers/mod*_dinov2_with_registers* @amyeroberts @qubvel
+/src/transformers/models/dit/mod*_dit* @amyeroberts @qubvel
+/src/transformers/models/dpt/mod*_dpt* @amyeroberts @qubvel
+/src/transformers/models/efficientformer/mod*_efficientformer* @amyeroberts @qubvel
+/src/transformers/models/efficientnet/mod*_efficientnet* @amyeroberts @qubvel
+/src/transformers/models/focalnet/mod*_focalnet* @amyeroberts @qubvel
+/src/transformers/models/glpn/mod*_glpn* @amyeroberts @qubvel
+/src/transformers/models/hiera/mod*_hiera* @amyeroberts @qubvel
+/src/transformers/models/ijepa/mod*_ijepa* @amyeroberts @qubvel
+/src/transformers/models/imagegpt/mod*_imagegpt* @amyeroberts @qubvel
+/src/transformers/models/levit/mod*_levit* @amyeroberts @qubvel
+/src/transformers/models/mask2former/mod*_mask2former* @amyeroberts @qubvel
+/src/transformers/models/maskformer/mod*_maskformer* @amyeroberts @qubvel
+/src/transformers/models/mobilenet_v1/mod*_mobilenet_v1* @amyeroberts @qubvel
+/src/transformers/models/mobilenet_v2/mod*_mobilenet_v2* @amyeroberts @qubvel
+/src/transformers/models/mobilevit/mod*_mobilevit* @amyeroberts @qubvel
+/src/transformers/models/mobilevitv2/mod*_mobilevitv2* @amyeroberts @qubvel
+/src/transformers/models/nat/mod*_nat* @amyeroberts @qubvel
+/src/transformers/models/poolformer/mod*_poolformer* @amyeroberts @qubvel
+/src/transformers/models/pvt/mod*_pvt* @amyeroberts @qubvel
+/src/transformers/models/pvt_v2/mod*_pvt_v2* @amyeroberts @qubvel
+/src/transformers/models/regnet/mod*_regnet* @amyeroberts @qubvel
+/src/transformers/models/resnet/mod*_resnet* @amyeroberts @qubvel
+/src/transformers/models/rt_detr/mod*_rt_detr* @amyeroberts @qubvel
+/src/transformers/models/segformer/mod*_segformer* @amyeroberts @qubvel
+/src/transformers/models/seggpt/mod*_seggpt* @amyeroberts @qubvel
+/src/transformers/models/superpoint/mod*_superpoint* @amyeroberts @qubvel
+/src/transformers/models/swiftformer/mod*_swiftformer* @amyeroberts @qubvel
+/src/transformers/models/swin/mod*_swin* @amyeroberts @qubvel
+/src/transformers/models/swinv2/mod*_swinv2* @amyeroberts @qubvel
+/src/transformers/models/swin2sr/mod*_swin2sr* @amyeroberts @qubvel
+/src/transformers/models/table_transformer/mod*_table_transformer* @amyeroberts @qubvel
+/src/transformers/models/textnet/mod*_textnet* @amyeroberts @qubvel
+/src/transformers/models/timm_wrapper/mod*_timm_wrapper* @amyeroberts @qubvel
+/src/transformers/models/upernet/mod*_upernet* @amyeroberts @qubvel
+/src/transformers/models/van/mod*_van* @amyeroberts @qubvel
+/src/transformers/models/vit/mod*_vit* @amyeroberts @qubvel
+/src/transformers/models/vit_hybrid/mod*_vit_hybrid* @amyeroberts @qubvel
+/src/transformers/models/vitdet/mod*_vitdet* @amyeroberts @qubvel
+/src/transformers/models/vit_mae/mod*_vit_mae* @amyeroberts @qubvel
+/src/transformers/models/vitmatte/mod*_vitmatte* @amyeroberts @qubvel
+/src/transformers/models/vit_msn/mod*_vit_msn* @amyeroberts @qubvel
+/src/transformers/models/vitpose/mod*_vitpose* @amyeroberts @qubvel
+/src/transformers/models/yolos/mod*_yolos* @amyeroberts @qubvel
+/src/transformers/models/zoedepth/mod*_zoedepth* @amyeroberts @qubvel
+
+# Audio models
+/src/transformers/models/audio_spectrogram_transformer/mod*_audio_spectrogram_transformer* @eustlb
+/src/transformers/models/bark/mod*_bark* @eustlb
+/src/transformers/models/clap/mod*_clap* @eustlb
+/src/transformers/models/dac/mod*_dac* @eustlb
+/src/transformers/models/encodec/mod*_encodec* @eustlb
+/src/transformers/models/hubert/mod*_hubert* @eustlb
+/src/transformers/models/mctct/mod*_mctct* @eustlb
+/src/transformers/models/mimi/mod*_mimi* @eustlb
+/src/transformers/models/mms/mod*_mms* @eustlb
+/src/transformers/models/moshi/mod*_moshi* @eustlb
+/src/transformers/models/musicgen/mod*_musicgen* @eustlb
+/src/transformers/models/musicgen_melody/mod*_musicgen_melody* @eustlb
+/src/transformers/models/pop2piano/mod*_pop2piano* @eustlb
+/src/transformers/models/seamless_m4t/mod*_seamless_m4t* @eustlb
+/src/transformers/models/seamless_m4t_v2/mod*_seamless_m4t_v2* @eustlb
+/src/transformers/models/sew/mod*_sew* @eustlb
+/src/transformers/models/sew_d/mod*_sew_d* @eustlb
+/src/transformers/models/speech_to_text/mod*_speech_to_text* @eustlb
+/src/transformers/models/speech_to_text_2/mod*_speech_to_text_2* @eustlb
+/src/transformers/models/speecht5/mod*_speecht5* @eustlb
+/src/transformers/models/unispeech/mod*_unispeech* @eustlb
+/src/transformers/models/unispeech_sat/mod*_unispeech_sat* @eustlb
+/src/transformers/models/univnet/mod*_univnet* @eustlb
+/src/transformers/models/vits/mod*_vits* @eustlb
+/src/transformers/models/wav2vec2/mod*_wav2vec2* @eustlb
+/src/transformers/models/wav2vec2_bert/mod*_wav2vec2_bert* @eustlb
+/src/transformers/models/wav2vec2_conformer/mod*_wav2vec2_conformer* @eustlb
+/src/transformers/models/wav2vec2_phoneme/mod*_wav2vec2_phoneme* @eustlb
+/src/transformers/models/wavlm/mod*_wavlm* @eustlb
+/src/transformers/models/whisper/mod*_whisper* @eustlb
+/src/transformers/models/xls_r/mod*_xls_r* @eustlb
+/src/transformers/models/xlsr_wav2vec2/mod*_xlsr_wav2vec2* @eustlb
+
+# Video models
+/src/transformers/models/timesformer/mod*_timesformer* @Rocketknight1
+/src/transformers/models/videomae/mod*_videomae* @Rocketknight1
+/src/transformers/models/vivit/mod*_vivit* @Rocketknight1
+
+# Multimodal models
+/src/transformers/models/align/mod*_align* @zucchini-nlp
+/src/transformers/models/altclip/mod*_altclip* @zucchini-nlp
+/src/transformers/models/aria/mod*_aria* @zucchini-nlp
+/src/transformers/models/blip/mod*_blip* @zucchini-nlp
+/src/transformers/models/blip_2/mod*_blip_2* @zucchini-nlp
+/src/transformers/models/bridgetower/mod*_bridgetower* @zucchini-nlp
+/src/transformers/models/bros/mod*_bros* @zucchini-nlp
+/src/transformers/models/chameleon/mod*_chameleon* @zucchini-nlp
+/src/transformers/models/chinese_clip/mod*_chinese_clip* @zucchini-nlp
+/src/transformers/models/clip/mod*_clip* @zucchini-nlp
+/src/transformers/models/clipseg/mod*_clipseg* @zucchini-nlp
+/src/transformers/models/clvp/mod*_clvp* @zucchini-nlp
+/src/transformers/models/colpali/mod*_colpali* @zucchini-nlp @yonigozlan
+/src/transformers/models/data2vec/mod*_data2vec* @zucchini-nlp
+/src/transformers/models/deplot/mod*_deplot* @zucchini-nlp
+/src/transformers/models/donut/mod*_donut* @zucchini-nlp
+/src/transformers/models/flava/mod*_flava* @zucchini-nlp
+/src/transformers/models/git/mod*_git* @zucchini-nlp
+/src/transformers/models/grounding_dino/mod*_grounding_dino* @qubvel
+/src/transformers/models/groupvit/mod*_groupvit* @zucchini-nlp
+/src/transformers/models/idefics/mod*_idefics* @zucchini-nlp
+/src/transformers/models/idefics2/mod*_idefics2* @zucchini-nlp
+/src/transformers/models/idefics3/mod*_idefics3* @zucchini-nlp
+/src/transformers/models/instructblip/mod*_instructblip* @zucchini-nlp
+/src/transformers/models/instructblipvideo/mod*_instructblipvideo* @zucchini-nlp
+/src/transformers/models/kosmos_2/mod*_kosmos_2* @zucchini-nlp
+/src/transformers/models/layoutlm/mod*_layoutlm* @NielsRogge
+/src/transformers/models/layoutlmv2/mod*_layoutlmv2* @NielsRogge
+/src/transformers/models/layoutlmv3/mod*_layoutlmv3* @NielsRogge
+/src/transformers/models/layoutxlm/mod*_layoutxlm* @NielsRogge
+/src/transformers/models/lilt/mod*_lilt* @zucchini-nlp
+/src/transformers/models/llava/mod*_llava* @zucchini-nlp @arthurzucker
+/src/transformers/models/llava_next/mod*_llava_next* @zucchini-nlp
+/src/transformers/models/llava_next_video/mod*_llava_next_video* @zucchini-nlp
+/src/transformers/models/llava_onevision/mod*_llava_onevision* @zucchini-nlp
+/src/transformers/models/lxmert/mod*_lxmert* @zucchini-nlp
+/src/transformers/models/matcha/mod*_matcha* @zucchini-nlp
+/src/transformers/models/mgp_str/mod*_mgp_str* @zucchini-nlp
+/src/transformers/models/mllama/mod*_mllama* @zucchini-nlp
+/src/transformers/models/nougat/mod*_nougat* @NielsRogge
+/src/transformers/models/omdet_turbo/mod*_omdet_turbo* @qubvel @yonigozlan
+/src/transformers/models/oneformer/mod*_oneformer* @zucchini-nlp
+/src/transformers/models/owlvit/mod*_owlvit* @qubvel
+/src/transformers/models/owlv2/mod*_owlv2* @qubvel
+/src/transformers/models/paligemma/mod*_paligemma* @zucchini-nlp @molbap
+/src/transformers/models/perceiver/mod*_perceiver* @zucchini-nlp
+/src/transformers/models/pix2struct/mod*_pix2struct* @zucchini-nlp
+/src/transformers/models/pixtral/mod*_pixtral* @zucchini-nlp @ArthurZucker
+/src/transformers/models/qwen2_audio/mod*_qwen2_audio* @zucchini-nlp @ArthurZucker
+/src/transformers/models/qwen2_vl/mod*_qwen2_vl* @zucchini-nlp @ArthurZucker
+/src/transformers/models/sam/mod*_sam* @zucchini-nlp @ArthurZucker
+/src/transformers/models/siglip/mod*_siglip* @zucchini-nlp
+/src/transformers/models/speech_encoder_decoder/mod*_speech_encoder_decoder* @zucchini-nlp
+/src/transformers/models/tapas/mod*_tapas* @NielsRogge
+/src/transformers/models/trocr/mod*_trocr* @zucchini-nlp
+/src/transformers/models/tvlt/mod*_tvlt* @zucchini-nlp
+/src/transformers/models/tvp/mod*_tvp* @zucchini-nlp
+/src/transformers/models/udop/mod*_udop* @zucchini-nlp
+/src/transformers/models/video_llava/mod*_video_llava* @zucchini-nlp
+/src/transformers/models/vilt/mod*_vilt* @zucchini-nlp
+/src/transformers/models/vipllava/mod*_vipllava* @zucchini-nlp
+/src/transformers/models/vision_encoder_decoder/mod*_vision_encoder_decoder* @Rocketknight1
+/src/transformers/models/vision_text_dual_encoder/mod*_vision_text_dual_encoder* @Rocketknight1
+/src/transformers/models/visual_bert/mod*_visual_bert* @zucchini-nlp
+/src/transformers/models/xclip/mod*_xclip* @zucchini-nlp
+
+# Reinforcement learning models
+/src/transformers/models/decision_transformer/mod*_decision_transformer* @Rocketknight1
+/src/transformers/models/trajectory_transformer/mod*_trajectory_transformer* @Rocketknight1
+
+# Time series models
+/src/transformers/models/autoformer/mod*_autoformer* @Rocketknight1
+/src/transformers/models/informer/mod*_informer* @Rocketknight1
+/src/transformers/models/patchtsmixer/mod*_patchtsmixer* @Rocketknight1
+/src/transformers/models/patchtst/mod*_patchtst* @Rocketknight1
+/src/transformers/models/time_series_transformer/mod*_time_series_transformer* @Rocketknight1
+
+# Graph models
+/src/transformers/models/graphormer/mod*_graphormer* @clefourrier
+
+# Finally, files with no owners that shouldn't generate pings, usually automatically generated and checked in the CI
+utils/dummy*

.github/workflows/assign-reviewers.yml

@@ -0,0 +1,26 @@
+name: Assign PR Reviewers
+on:
+  pull_request_target:
+    branches:
+      - main
+    types: [ready_for_review]
+
+jobs:
+  assign_reviewers:
+    permissions:
+       pull-requests: write
+    runs-on: ubuntu-22.04
+    steps:
+      - uses: actions/checkout@v4
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.13'
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip
+          pip install PyGithub
+      - name: Run assignment script
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        run: python .github/scripts/assign_reviewers.py

.github/workflows/build-docker-images.yml

@@ -63,14 +63,14 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-all-latest-gpu-push-ci docker build 
+          title: 🤗 Results of the transformers-all-latest-gpu-push-ci docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
   latest-torch-deepspeed-docker:
     name: "Latest PyTorch + DeepSpeed"
     runs-on:
-      group: aws-general-8-plus
+      group: aws-g4dn-2xlarge-cache
     steps:
       -
         name: Set up Docker Buildx
@@ -99,7 +99,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER}}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu docker build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -140,7 +140,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu-push-ci docker build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-latest-gpu-push-ci docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -176,7 +176,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-doc-builder docker build 
+          title: 🤗 Results of the huggingface/transformers-doc-builder docker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -214,7 +214,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-gpudocker build 
+          title: 🤗 Results of the huggingface/transformers-pytorch-gpudocker build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -223,19 +223,19 @@ jobs:
     runs-on:
       group: aws-general-8-plus
     steps:
-      - 
+      -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
-      - 
+      -
         name: Check out code
         uses: actions/checkout@v4
-      - 
+      -
         name: Login to DockerHub
         uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
+      -
         name: Build and push
         uses: docker/build-push-action@v5
         with:
@@ -263,7 +263,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-pytorch-amd-gpu-push-ci build 
+          title: 🤗 Results of the huggingface/transformers-pytorch-amd-gpu-push-ci build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -301,7 +301,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the huggingface/transformers-tensorflow-gpu build 
+          title: 🤗 Results of the huggingface/transformers-tensorflow-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -310,19 +310,19 @@ jobs:
     runs-on:
       group: aws-general-8-plus
     steps:
-      - 
+      -
         name: Set up Docker Buildx
         uses: docker/setup-buildx-action@v3
-      - 
+      -
         name: Check out code
         uses: actions/checkout@v4
-      - 
+      -
         name: Login to DockerHub
         uses: docker/login-action@v3
         with:
           username: ${{ secrets.DOCKERHUB_USERNAME }}
           password: ${{ secrets.DOCKERHUB_PASSWORD }}
-      - 
+      -
         name: Build and push
         uses: docker/build-push-action@v5
         with:
@@ -350,7 +350,7 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-pytorch-deepspeed-amd-gpu build 
+          title: 🤗 Results of the transformers-pytorch-deepspeed-amd-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}
 
@@ -388,6 +388,6 @@ jobs:
         uses: huggingface/hf-workflows/.github/actions/post-slack@main
         with:
           slack_channel: ${{ secrets.CI_SLACK_CHANNEL_DOCKER }}
-          title: 🤗 Results of the transformers-quantization-latest-gpu build 
+          title: 🤗 Results of the transformers-quantization-latest-gpu build
           status: ${{ job.status }}
           slack_token: ${{ secrets.SLACK_CIFEEDBACK_BOT_TOKEN }}

.github/workflows/build-nightly-ci-docker-images.yml

@@ -42,7 +42,7 @@ jobs:
   nightly-torch-deepspeed-docker:
     name: "Nightly PyTorch + DeepSpeed"
     runs-on:
-      group: aws-general-8-plus
+      group: aws-g4dn-2xlarge-cache
     steps:
       -
         name: Set up Docker Buildx

.github/workflows/build_pr_documentation.yml

@@ -14,5 +14,4 @@ jobs:
       commit_sha: ${{ github.event.pull_request.head.sha }}
       pr_number: ${{ github.event.number }}
       package: transformers
-      languages: ar de en es fr hi it ko pt tr zh ja te
-      custom_container: huggingface/transformers-doc-builder
+      languages: en

.github/workflows/change_pr_to_draft.yml

@@ -22,4 +22,4 @@ jobs:
         run: |
           echo $PR_NUMBER
           gh pr ready $PR_NUMBER --repo $REPO --undo
-          gh pr comment $PR_NUMBER --repo $REPO --body "Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. When it is ready for review, please click the \`Ready for review\` button (at the bottom of the PR page)."
+          gh pr comment $PR_NUMBER --repo $REPO --body "Hi 👋, thank you for opening this pull request! The pull request is converted to draft by default. The CI will be paused while the PR is in draft mode. When it is ready for review, please click the \`Ready for review\` button (at the bottom of the PR page). This will assign reviewers and trigger CI."

.github/workflows/model_jobs.yml

@@ -18,6 +18,10 @@ on:
       docker:
         required: true
         type: string
+      report_name_prefix:
+        required: false
+        default: run_models_gpu
+        type: string
 
 env:
   HF_HOME: /mnt/cache
@@ -116,23 +120,23 @@ jobs:
 
       - name: Run all tests on GPU
         working-directory: /transformers
-        run: python3 -m pytest -rsfE -v --make-reports=${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
+        run: python3 -m pytest -rsfE -v --make-reports=${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports tests/${{ matrix.folders }}
 
       - name: Failure short reports
         if: ${{ failure() }}
         continue-on-error: true
-        run: cat /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/failures_short.txt
+        run: cat /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports/failures_short.txt
 
       - name: Run test
         shell: bash
         run: |
-          mkdir -p /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
-          echo "hello" > /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports/hello.txt
-          echo "${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports"
+          mkdir -p /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports
+          echo "hello" > /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports/hello.txt
+          echo "${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports"
 
-      - name: "Test suite reports artifacts: ${{ env.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports"
+      - name: "Test suite reports artifacts: ${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ env.matrix_folders }}_test_reports"
         if: ${{ always() }}
         uses: actions/upload-artifact@v4
         with:
-          name: ${{ env.machine_type }}_run_models_gpu_${{ env.matrix_folders }}_test_reports
-          path: /transformers/reports/${{ env.machine_type }}_run_models_gpu_${{ matrix.folders }}_test_reports
+          name: ${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ env.matrix_folders }}_test_reports
+          path: /transformers/reports/${{ env.machine_type }}_${{ inputs.report_name_prefix }}_${{ matrix.folders }}_test_reports

.github/workflows/push-important-models.yml

@@ -27,7 +27,7 @@ jobs:
 
       - name: Get changed files
         id: changed-files
-        uses: tj-actions/changed-files@3f54ebb830831fc121d3263c1857cfbdc310cdb9 #v42
+        uses: tj-actions/changed-files@1c8e6069583811afb28f97afeaf8e7da80c6be5c
         with:
           files: src/transformers/models/**
 

.github/workflows/self-comment-ci.yml

@@ -29,7 +29,7 @@ jobs:
     runs-on: ubuntu-22.04
     name: Get PR number
     # For security: only allow team members to run
-    if: ${{ github.event.issue.state == 'open' && contains(fromJSON('["ydshieh", "ArthurZucker", "zucchini-nlp", "qubvel", "molbap", "gante", "LysandreJik", "Cyrilvallez", "Rocketknight1", "SunMarc", "muellerzr"]'), github.actor) && (startsWith(github.event.comment.body, 'run-slow') || startsWith(github.event.comment.body, 'run slow') || startsWith(github.event.comment.body, 'run_slow')) }}
+    if: ${{ github.event.issue.state == 'open' && contains(fromJSON('["ydshieh", "ArthurZucker", "zucchini-nlp", "qubvel", "molbap", "gante", "LysandreJik", "Cyrilvallez", "Rocketknight1", "SunMarc", "muellerzr", "eustlb", "MekkCyber"]'), github.actor) && (startsWith(github.event.comment.body, 'run-slow') || startsWith(github.event.comment.body, 'run slow') || startsWith(github.event.comment.body, 'run_slow')) }}
     outputs:
       PR_NUMBER: ${{ steps.set_pr_number.outputs.PR_NUMBER }}
     steps:

.github/workflows/self-push-caller.yml

@@ -25,7 +25,7 @@ jobs:
         
         - name: Get changed files
           id: changed-files
-          uses: tj-actions/changed-files@v41
+          uses: tj-actions/changed-files@1c8e6069583811afb28f97afeaf8e7da80c6be5c
         
         - name: Was setup changed 
           id: was_changed
@@ -51,4 +51,4 @@ jobs:
     needs: build-docker-containers
     steps:
       - name: Trigger push CI via workflow_run
-        run: echo "Trigger push CI via workflow_run"
+        run: echo "Trigger push CI via workflow_run"

.github/workflows/self-scheduled-caller.yml

@@ -54,12 +54,23 @@ jobs:
       ci_event: Daily CI
     secrets: inherit
 
+  trainer-fsdp-ci:
+    name: Trainer/FSDP CI
+    uses: ./.github/workflows/self-scheduled.yml
+    with:
+      job: run_trainer_and_fsdp_gpu
+      slack_report_channel: "#transformers-ci-daily-training"
+      runner: daily-ci
+      docker: huggingface/transformers-all-latest-gpu
+      ci_event: Daily CI
+    secrets: inherit
+
   deepspeed-ci:
     name: DeepSpeed CI
     uses: ./.github/workflows/self-scheduled.yml
     with:
       job: run_torch_cuda_extensions_gpu
-      slack_report_channel: "#transformers-ci-daily-deepspeed"
+      slack_report_channel: "#transformers-ci-daily-training"
       runner: daily-ci
       docker: huggingface/transformers-pytorch-deepspeed-latest-gpu
       ci_event: Daily CI

.github/workflows/self-scheduled.yml

@@ -45,7 +45,7 @@ env:
 
 jobs:
   setup:
-    if: contains(fromJSON('["run_models_gpu", "run_quantization_torch_gpu"]'), inputs.job)
+    if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu", "run_quantization_torch_gpu"]'), inputs.job)
     name: Setup
     strategy:
       matrix:
@@ -77,12 +77,17 @@ jobs:
         run: pip freeze
 
       - id: set-matrix
-        if: ${{ inputs.job == 'run_models_gpu' }}
+        if: contains(fromJSON('["run_models_gpu", "run_trainer_and_fsdp_gpu"]'), inputs.job)
         name: Identify models to test
         working-directory: /transformers/tests
         run: |
-          echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
-          echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
+          if [ "${{ inputs.job }}" = "run_models_gpu" ]; then
+            echo "folder_slices=$(python3 ../utils/split_model_tests.py --num_splits ${{ env.NUM_SLICES }})" >> $GITHUB_OUTPUT
+            echo "slice_ids=$(python3 -c 'd = list(range(${{ env.NUM_SLICES }})); print(d)')" >> $GITHUB_OUTPUT
+          elif [ "${{ inputs.job }}" = "run_trainer_and_fsdp_gpu" ]; then
+            echo "folder_slices=[['trainer'], ['fsdp']]" >> $GITHUB_OUTPUT
+            echo "slice_ids=[0, 1]" >> $GITHUB_OUTPUT
+          fi
 
       - id: set-matrix-quantization
         if: ${{ inputs.job == 'run_quantization_torch_gpu' }}
@@ -113,6 +118,25 @@ jobs:
       docker: ${{ inputs.docker }}
     secrets: inherit
 
+  run_trainer_and_fsdp_gpu:
+    if: ${{ inputs.job == 'run_trainer_and_fsdp_gpu' }}
+    name: " "
+    needs: setup
+    strategy:
+      fail-fast: false
+      matrix:
+        machine_type: [aws-g4dn-2xlarge-cache, aws-g4dn-12xlarge-cache]
+        slice_id: [0, 1]
+    uses: ./.github/workflows/model_jobs.yml
+    with:
+      folder_slices: ${{ needs.setup.outputs.folder_slices }}
+      machine_type: ${{ matrix.machine_type }}
+      slice_id: ${{ matrix.slice_id }}
+      runner: ${{ inputs.runner }}
+      docker: ${{ inputs.docker }}
+      report_name_prefix: run_trainer_and_fsdp_gpu
+    secrets: inherit
+
   run_pipelines_torch_gpu:
     if: ${{ inputs.job == 'run_pipelines_torch_gpu' }}
     name: PyTorch pipelines
@@ -382,7 +406,7 @@ jobs:
         run: pip freeze
 
       - name: Set `machine_type` for report and artifact names
-        working-directory: /transformers
+        working-directory: ${{ inputs.working-directory-prefix }}/transformers
         shell: bash
         run: |
           echo "${{ matrix.machine_type }}"
@@ -541,6 +565,7 @@ jobs:
     needs: [
       setup,
       run_models_gpu,
+      run_trainer_and_fsdp_gpu,
       run_pipelines_torch_gpu,
       run_pipelines_tf_gpu,
       run_examples_gpu,

.github/workflows/update_metdata.yml

@@ -19,7 +19,7 @@ jobs:
       - name: Setup environment
         run: |
           pip install --upgrade pip
-          pip install datasets pandas==2.0.3
+          pip install datasets pandas
           pip install .[torch,tf,flax]
 
       - name: Update metadata

CONTRIBUTING.md

@@ -221,10 +221,10 @@ You'll need **[Python 3.9](https://github.com/huggingface/transformers/blob/main
    [Checks on a Pull Request](https://huggingface.co/docs/transformers/pr_checks) guide.
 
    If you're modifying documents under the `docs/source` directory, make sure the documentation can still be built. This check will also run in the CI when you open a pull request. To run a local check
-   make sure you install the documentation builder:
+   make sure you install the [documentation builder](https://github.com/huggingface/doc-builder).
 
    ```bash
-   pip install ".[docs]"
+   pip install hf-doc-builder
    ```
 
    Run the following command from the root of the repository:

ISSUES.md

@@ -26,7 +26,7 @@ There are two main venues to receive support: [the forums](https://discuss.huggi
 
 [The user forums](https://discuss.huggingface.co/) are supported by the wide community of the library users and backed up by developers when needed.
 
-If you have a difficulty with deploying this library or some questions, or you'd like to discuss a new feature, please first consider discussing those things at the forums. Only when you feel your subject matter has been crystalized and you still need support from the library developers do proceed to file an [issue](https://github.com/huggingface/transformers/issues).
+If you have a difficulty with deploying this library or some questions, or you'd like to discuss a new feature, please first consider discussing those things at the forums. Only when you feel your subject matter has been crystallized and you still need support from the library developers do proceed to file an [issue](https://github.com/huggingface/transformers/issues).
 
 In particular all "Please explain" questions or objectively very user-specific feature requests belong to the forums. Here are some example of such questions:
 
@@ -263,9 +263,9 @@ You are not required to read the following guidelines before opening an issue. H
     But if you're replying to a comment that happened some comments back it's always a good practice to quote just the relevant lines you're replying it. The `>` is used for quoting, or you can always use the menu to do so. For example your editor box will look like:
 
     ```
-    > How big is your gpu cluster?
+    > How big is your GPU cluster?
 
-    Our cluster is made of 256 gpus.
+    Our cluster is made of 256 GPUs.
     ```
 
     If you are addressing multiple comments, quote the relevant parts of each before your answer. Some people use the same comment to do multiple replies, others separate them into separate comments. Either way works. The latter approach helps for linking to a specific comment.

README.md

@@ -25,6 +25,7 @@ limitations under the License.
 </p>
 
 <p align="center">
+    <a href="https://huggingface.com/models"><img alt="Checkpoints on Hub" src="https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen"></a>
     <a href="https://circleci.com/gh/huggingface/transformers"><img alt="Build" src="https://img.shields.io/circleci/build/github/huggingface/transformers/main"></a>
     <a href="https://github.com/huggingface/transformers/blob/main/LICENSE"><img alt="GitHub" src="https://img.shields.io/github/license/huggingface/transformers.svg?color=blue"></a>
     <a href="https://huggingface.co/docs/transformers/index"><img alt="Documentation" src="https://img.shields.io/website/http/huggingface.co/docs/transformers/index.svg?down_color=red&down_message=offline&up_message=online"></a>
@@ -54,275 +55,254 @@ limitations under the License.
 </h4>
 
 <h3 align="center">
-    <p>State-of-the-art Machine Learning for JAX, PyTorch and TensorFlow</p>
+    <p>State-of-the-art pretrained models for inference and training</p>
 </h3>
 
 <h3 align="center">
     <a href="https://hf.co/course"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/course_banner.png"></a>
 </h3>
 
-🤗 Transformers provides thousands of pretrained models to perform tasks on different modalities such as text, vision, and audio.
+Transformers is a library of pretrained text, computer vision, audio, video, and multimodal models for inference and training. Use Transformers to fine-tune models on your data, build inference applications, and for generative AI use cases across multiple modalities.
 
-These models can be applied on:
+There are over 500K+ Transformers [model checkpoints](https://huggingface.co/models?library=transformers&sort=trending) on the [Hugging Face Hub](https://huggingface.com/models) you can use.
 
-* 📝 Text, for tasks like text classification, information extraction, question answering, summarization, translation, and text generation, in over 100 languages.
-* 🖼️ Images, for tasks like image classification, object detection, and segmentation.
-* 🗣️ Audio, for tasks like speech recognition and audio classification.
+Explore the [Hub](https://huggingface.com/) today to find a model and use Transformers to help you get started right away.
 
-Transformer models can also perform tasks on **several modalities combined**, such as table question answering, optical character recognition, information extraction from scanned documents, video classification, and visual question answering.
+## Installation
 
-🤗 Transformers provides APIs to quickly download and use those pretrained models on a given text, fine-tune them on your own datasets and then share them with the community on our [model hub](https://huggingface.co/models). At the same time, each python module defining an architecture is fully standalone and can be modified to enable quick research experiments.
+Transformers works with Python 3.9+ [PyTorch](https://pytorch.org/get-started/locally/) 2.1+, [TensorFlow](https://www.tensorflow.org/install/pip) 2.6+, and [Flax](https://flax.readthedocs.io/en/latest/) 0.4.1+.
 
-🤗 Transformers is backed by the three most popular deep learning libraries — [Jax](https://jax.readthedocs.io/en/latest/), [PyTorch](https://pytorch.org/) and [TensorFlow](https://www.tensorflow.org/) — with a seamless integration between them. It's straightforward to train your models with one before loading them for inference with the other.
+Create and activate a virtual environment with [venv](https://docs.python.org/3/library/venv.html) or [uv](https://docs.astral.sh/uv/), a fast Rust-based Python package and project manager.
 
-## Online demos
+```py
+# venv
+python -m venv .my-env
+source .my-env/bin/activate
 
-You can test most of our models directly on their pages from the [model hub](https://huggingface.co/models). We also offer [private model hosting, versioning, & an inference API](https://huggingface.co/pricing) for public and private models.
-
-Here are a few examples:
-
-In Natural Language Processing:
-- [Masked word completion with BERT](https://huggingface.co/google-bert/bert-base-uncased?text=Paris+is+the+%5BMASK%5D+of+France)
-- [Named Entity Recognition with Electra](https://huggingface.co/dbmdz/electra-large-discriminator-finetuned-conll03-english?text=My+name+is+Sarah+and+I+live+in+London+city)
-- [Text generation with Mistral](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2)
-- [Natural Language Inference with RoBERTa](https://huggingface.co/FacebookAI/roberta-large-mnli?text=The+dog+was+lost.+Nobody+lost+any+animal)
-- [Summarization with BART](https://huggingface.co/facebook/bart-large-cnn?text=The+tower+is+324+metres+%281%2C063+ft%29+tall%2C+about+the+same+height+as+an+81-storey+building%2C+and+the+tallest+structure+in+Paris.+Its+base+is+square%2C+measuring+125+metres+%28410+ft%29+on+each+side.+During+its+construction%2C+the+Eiffel+Tower+surpassed+the+Washington+Monument+to+become+the+tallest+man-made+structure+in+the+world%2C+a+title+it+held+for+41+years+until+the+Chrysler+Building+in+New+York+City+was+finished+in+1930.+It+was+the+first+structure+to+reach+a+height+of+300+metres.+Due+to+the+addition+of+a+broadcasting+aerial+at+the+top+of+the+tower+in+1957%2C+it+is+now+taller+than+the+Chrysler+Building+by+5.2+metres+%2817+ft%29.+Excluding+transmitters%2C+the+Eiffel+Tower+is+the+second+tallest+free-standing+structure+in+France+after+the+Millau+Viaduct)
-- [Question answering with DistilBERT](https://huggingface.co/distilbert/distilbert-base-uncased-distilled-squad?text=Which+name+is+also+used+to+describe+the+Amazon+rainforest+in+English%3F&context=The+Amazon+rainforest+%28Portuguese%3A+Floresta+Amaz%C3%B4nica+or+Amaz%C3%B4nia%3B+Spanish%3A+Selva+Amaz%C3%B3nica%2C+Amazon%C3%ADa+or+usually+Amazonia%3B+French%3A+For%C3%AAt+amazonienne%3B+Dutch%3A+Amazoneregenwoud%29%2C+also+known+in+English+as+Amazonia+or+the+Amazon+Jungle%2C+is+a+moist+broadleaf+forest+that+covers+most+of+the+Amazon+basin+of+South+America.+This+basin+encompasses+7%2C000%2C000+square+kilometres+%282%2C700%2C000+sq+mi%29%2C+of+which+5%2C500%2C000+square+kilometres+%282%2C100%2C000+sq+mi%29+are+covered+by+the+rainforest.+This+region+includes+territory+belonging+to+nine+nations.+The+majority+of+the+forest+is+contained+within+Brazil%2C+with+60%25+of+the+rainforest%2C+followed+by+Peru+with+13%25%2C+Colombia+with+10%25%2C+and+with+minor+amounts+in+Venezuela%2C+Ecuador%2C+Bolivia%2C+Guyana%2C+Suriname+and+French+Guiana.+States+or+departments+in+four+nations+contain+%22Amazonas%22+in+their+names.+The+Amazon+represents+over+half+of+the+planet%27s+remaining+rainforests%2C+and+comprises+the+largest+and+most+biodiverse+tract+of+tropical+rainforest+in+the+world%2C+with+an+estimated+390+billion+individual+trees+divided+into+16%2C000+species)
-- [Translation with T5](https://huggingface.co/google-t5/t5-base?text=My+name+is+Wolfgang+and+I+live+in+Berlin)
-
-In Computer Vision:
-- [Image classification with ViT](https://huggingface.co/google/vit-base-patch16-224)
-- [Object Detection with DETR](https://huggingface.co/facebook/detr-resnet-50)
-- [Semantic Segmentation with SegFormer](https://huggingface.co/nvidia/segformer-b0-finetuned-ade-512-512)
-- [Panoptic Segmentation with Mask2Former](https://huggingface.co/facebook/mask2former-swin-large-coco-panoptic)
-- [Depth Estimation with Depth Anything](https://huggingface.co/docs/transformers/main/model_doc/depth_anything)
-- [Video Classification with VideoMAE](https://huggingface.co/docs/transformers/model_doc/videomae)
-- [Universal Segmentation with OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_dinat_large)
-
-In Audio:
-- [Automatic Speech Recognition with Whisper](https://huggingface.co/openai/whisper-large-v3)
-- [Keyword Spotting with Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
-- [Audio Classification with Audio Spectrogram Transformer](https://huggingface.co/MIT/ast-finetuned-audioset-10-10-0.4593)
-
-In Multimodal tasks:
-- [Table Question Answering with TAPAS](https://huggingface.co/google/tapas-base-finetuned-wtq)
-- [Visual Question Answering with ViLT](https://huggingface.co/dandelin/vilt-b32-finetuned-vqa)
-- [Image captioning with LLaVa](https://huggingface.co/llava-hf/llava-1.5-7b-hf)
-- [Zero-shot Image Classification with SigLIP](https://huggingface.co/google/siglip-so400m-patch14-384)
-- [Document Question Answering with LayoutLM](https://huggingface.co/impira/layoutlm-document-qa)
-- [Zero-shot Video Classification with X-CLIP](https://huggingface.co/docs/transformers/model_doc/xclip)
-- [Zero-shot Object Detection with OWLv2](https://huggingface.co/docs/transformers/en/model_doc/owlv2)
-- [Zero-shot Image Segmentation with CLIPSeg](https://huggingface.co/docs/transformers/model_doc/clipseg)
-- [Automatic Mask Generation with SAM](https://huggingface.co/docs/transformers/model_doc/sam)
-
-
-## 100 projects using Transformers
-
-Transformers is more than a toolkit to use pretrained models: it's a community of projects built around it and the
-Hugging Face Hub. We want Transformers to enable developers, researchers, students, professors, engineers, and anyone
-else to build their dream projects.
-
-In order to celebrate the 100,000 stars of transformers, we have decided to put the spotlight on the
-community, and we have created the [awesome-transformers](./awesome-transformers.md) page which lists 100
-incredible projects built in the vicinity of transformers.
-
-If you own or use a project that you believe should be part of the list, please open a PR to add it!
-
-## Serious about AI in your organisation? Build faster with the Hugging Face Enterprise Hub.
-
-<a target="_blank" href="https://huggingface.co/enterprise">
-    <img alt="Hugging Face Enterprise Hub" src="https://github.com/user-attachments/assets/247fb16d-d251-4583-96c4-d3d76dda4925">
-</a><br>
-
-## Quick tour
-
-To immediately use a model on a given input (text, image, audio, ...), we provide the `pipeline` API. Pipelines group together a pretrained model with the preprocessing that was used during that model's training. Here is how to quickly use a pipeline to classify positive versus negative texts:
-
-```python
->>> from transformers import pipeline
-
-# Allocate a pipeline for sentiment-analysis
->>> classifier = pipeline('sentiment-analysis')
->>> classifier('We are very happy to introduce pipeline to the transformers repository.')
-[{'label': 'POSITIVE', 'score': 0.9996980428695679}]
+# uv
+uv venv .my-env
+source .my-env/bin/activate
 ```
 
-The second line of code downloads and caches the pretrained model used by the pipeline, while the third evaluates it on the given text. Here, the answer is "positive" with a confidence of 99.97%.
+Install Transformers in your virtual environment.
 
-Many tasks have a pre-trained `pipeline` ready to go, in NLP but also in computer vision and speech. For example, we can easily extract detected objects in an image:
+```py
+# pip
+pip install transformers
 
-``` python
->>> import requests
->>> from PIL import Image
->>> from transformers import pipeline
-
-# Download an image with cute cats
->>> url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png"
->>> image_data = requests.get(url, stream=True).raw
->>> image = Image.open(image_data)
-
-# Allocate a pipeline for object detection
->>> object_detector = pipeline('object-detection')
->>> object_detector(image)
-[{'score': 0.9982201457023621,
-  'label': 'remote',
-  'box': {'xmin': 40, 'ymin': 70, 'xmax': 175, 'ymax': 117}},
- {'score': 0.9960021376609802,
-  'label': 'remote',
-  'box': {'xmin': 333, 'ymin': 72, 'xmax': 368, 'ymax': 187}},
- {'score': 0.9954745173454285,
-  'label': 'couch',
-  'box': {'xmin': 0, 'ymin': 1, 'xmax': 639, 'ymax': 473}},
- {'score': 0.9988006353378296,
-  'label': 'cat',
-  'box': {'xmin': 13, 'ymin': 52, 'xmax': 314, 'ymax': 470}},
- {'score': 0.9986783862113953,
-  'label': 'cat',
-  'box': {'xmin': 345, 'ymin': 23, 'xmax': 640, 'ymax': 368}}]
+# uv
+uv pip install transformers
 ```
 
-Here, we get a list of objects detected in the image, with a box surrounding the object and a confidence score. Here is the original image on the left, with the predictions displayed on the right:
+Install Transformers from source if you want the latest changes in the library or are interested in contributing. However, the *latest* version may not be stable. Feel free to open an [issue](https://github.com/huggingface/transformers/issues) if you encounter an error.
+
+```shell
+git clone https://github.com/huggingface/transformers.git
+cd transformers
+pip install .
+```
+
+## Quickstart
+
+Get started with Transformers right away with the [Pipeline](https://huggingface.co/docs/transformers/pipeline_tutorial) API. The `Pipeline` is a high-level inference class that supports text, audio, vision, and multimodal tasks. It handles preprocessing the input and returns the appropriate output.
+
+Instantiate a pipeline and specify model to use for text generation. The model is downloaded and cached so you can easily reuse it again. Finally, pass some text to prompt the model.
+
+```py
+from transformers import pipeline
+
+pipeline = pipeline(task="text-generation", model="Qwen/Qwen2.5-1.5B")
+pipeline("the secret to baking a really good cake is ")
+[{'generated_text': 'the secret to baking a really good cake is 1) to use the right ingredients and 2) to follow the recipe exactly. the recipe for the cake is as follows: 1 cup of sugar, 1 cup of flour, 1 cup of milk, 1 cup of butter, 1 cup of eggs, 1 cup of chocolate chips. if you want to make 2 cakes, how much sugar do you need? To make 2 cakes, you will need 2 cups of sugar.'}]
+```
+
+To chat with a model, the usage pattern is the same. The only difference is you need to construct a chat history (the input to `Pipeline`) between you and the system.
+
+> [!TIP]
+> You can also chat with a model directly from the command line.
+> ```shell
+> transformers-cli chat --model_name_or_path Qwen/Qwen2.5-0.5B-Instruct
+> ```
+
+```py
+import torch
+from transformers import pipeline
+
+chat = [
+    {"role": "system", "content": "You are a sassy, wise-cracking robot as imagined by Hollywood circa 1986."},
+    {"role": "user", "content": "Hey, can you tell me any fun things to do in New York?"}
+]
+
+pipeline = pipeline(task="text-generation", model="meta-llama/Meta-Llama-3-8B-Instruct", torch_dtype=torch.bfloat16, device_map="auto")
+response = pipeline(chat, max_new_tokens=512)
+print(response[0]["generated_text"][-1]["content"])
+```
+
+Expand the examples below to see how `Pipeline` works for different modalities and tasks.
+
+<details>
+<summary>Automatic speech recognition</summary>
+
+```py
+from transformers import pipeline
+
+pipeline = pipeline(task="automatic-speech-recognition", model="openai/whisper-large-v3")
+pipeline("https://huggingface.co/datasets/Narsil/asr_dummy/resolve/main/mlk.flac")
+{'text': ' I have a dream that one day this nation will rise up and live out the true meaning of its creed.'}
+```
+
+</details>
+
+<details>
+<summary>Image classification</summary>
 
 <h3 align="center">
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample.png" width="400"></a>
-    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/coco_sample_post_processed.png" width="400"></a>
+    <a><img src="https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png"></a>
 </h3>
 
-You can learn more about the tasks supported by the `pipeline` API in [this tutorial](https://huggingface.co/docs/transformers/task_summary).
+```py
+from transformers import pipeline
 
-In addition to `pipeline`, to download and use any of the pretrained models on your given task, all it takes is three lines of code. Here is the PyTorch version:
-```python
->>> from transformers import AutoTokenizer, AutoModel
-
->>> tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
->>> model = AutoModel.from_pretrained("google-bert/bert-base-uncased")
-
->>> inputs = tokenizer("Hello world!", return_tensors="pt")
->>> outputs = model(**inputs)
+pipeline = pipeline(task="image-classification", model="facebook/dinov2-small-imagenet1k-1-layer")
+pipeline("https://huggingface.co/datasets/Narsil/image_dummy/raw/main/parrots.png")
+[{'label': 'macaw', 'score': 0.997848391532898},
+ {'label': 'sulphur-crested cockatoo, Kakatoe galerita, Cacatua galerita',
+  'score': 0.0016551691805943847},
+ {'label': 'lorikeet', 'score': 0.00018523589824326336},
+ {'label': 'African grey, African gray, Psittacus erithacus',
+  'score': 7.85409429227002e-05},
+ {'label': 'quail', 'score': 5.502637941390276e-05}]
 ```
 
-And here is the equivalent code for TensorFlow:
-```python
->>> from transformers import AutoTokenizer, TFAutoModel
+</details>
 
->>> tokenizer = AutoTokenizer.from_pretrained("google-bert/bert-base-uncased")
->>> model = TFAutoModel.from_pretrained("google-bert/bert-base-uncased")
+<details>
+<summary>Visual question answering</summary>
 
->>> inputs = tokenizer("Hello world!", return_tensors="tf")
->>> outputs = model(**inputs)
+
+<h3 align="center">
+    <a><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-few-shot.jpg"></a>
+</h3>
+
+```py
+from transformers import pipeline
+
+pipeline = pipeline(task="visual-question-answering", model="Salesforce/blip-vqa-base")
+pipeline(
+    image="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/idefics-few-shot.jpg",
+    question="What is in the image?",
+)
+[{'answer': 'statue of liberty'}]
 ```
 
-The tokenizer is responsible for all the preprocessing the pretrained model expects and can be called directly on a single string (as in the above examples) or a list. It will output a dictionary that you can use in downstream code or simply directly pass to your model using the ** argument unpacking operator.
+</details>
 
-The model itself is a regular [Pytorch `nn.Module`](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) or a [TensorFlow `tf.keras.Model`](https://www.tensorflow.org/api_docs/python/tf/keras/Model) (depending on your backend) which you can use as usual. [This tutorial](https://huggingface.co/docs/transformers/training) explains how to integrate such a model into a classic PyTorch or TensorFlow training loop, or how to use our `Trainer` API to quickly fine-tune on a new dataset.
-
-## Why should I use transformers?
+## Why should I use Transformers?
 
 1. Easy-to-use state-of-the-art models:
-    - High performance on natural language understanding & generation, computer vision, and audio tasks.
-    - Low barrier to entry for educators and practitioners.
+    - High performance on natural language understanding & generation, computer vision, audio, video, and multimodal tasks.
+    - Low barrier to entry for researchers, engineers, and developers.
     - Few user-facing abstractions with just three classes to learn.
     - A unified API for using all our pretrained models.
 
 1. Lower compute costs, smaller carbon footprint:
-    - Researchers can share trained models instead of always retraining.
-    - Practitioners can reduce compute time and production costs.
-    - Dozens of architectures with over 400,000 pretrained models across all modalities.
+    - Share trained models instead of training from scratch.
+    - Reduce compute time and production costs.
+    - Dozens of model architectures with 1M+ pretrained checkpoints across all modalities.
 
-1. Choose the right framework for every part of a model's lifetime:
+1. Choose the right framework for every part of a models lifetime:
     - Train state-of-the-art models in 3 lines of code.
-    - Move a single model between TF2.0/PyTorch/JAX frameworks at will.
-    - Seamlessly pick the right framework for training, evaluation, and production.
+    - Move a single model between PyTorch/JAX/TF2.0 frameworks at will.
+    - Pick the right framework for training, evaluation, and production.
 
 1. Easily customize a model or an example to your needs:
     - We provide examples for each architecture to reproduce the results published by its original authors.
     - Model internals are exposed as consistently as possible.
     - Model files can be used independently of the library for quick experiments.
 
-## Why shouldn't I use transformers?
+<a target="_blank" href="https://huggingface.co/enterprise">
+    <img alt="Hugging Face Enterprise Hub" src="https://github.com/user-attachments/assets/247fb16d-d251-4583-96c4-d3d76dda4925">
+</a><br>
+
+## Why shouldn't I use Transformers?
 
 - This library is not a modular toolbox of building blocks for neural nets. The code in the model files is not refactored with additional abstractions on purpose, so that researchers can quickly iterate on each of the models without diving into additional abstractions/files.
-- The training API is not intended to work on any model but is optimized to work with the models provided by the library. For generic machine learning loops, you should use another library (possibly, [Accelerate](https://huggingface.co/docs/accelerate)).
-- While we strive to present as many use cases as possible, the scripts in our [examples folder](https://github.com/huggingface/transformers/tree/main/examples) are just that: examples. It is expected that they won't work out-of-the-box on your specific problem and that you will be required to change a few lines of code to adapt them to your needs.
+- The training API is optimized to work with PyTorch models provided by Transformers. For generic machine learning loops, you should use another library like [Accelerate](https://huggingface.co/docs/accelerate).
+- The [example scripts]((https://github.com/huggingface/transformers/tree/main/examples)) are only *examples*. They may not necessarily work out-of-the-box on your specific use case and you'll need to adapt the code for it to work.
 
-## Installation
+## 100 projects using Transformers
 
-### With pip
+Transformers is more than a toolkit to use pretrained models, it's a community of projects built around it and the
+Hugging Face Hub. We want Transformers to enable developers, researchers, students, professors, engineers, and anyone
+else to build their dream projects.
 
-This repository is tested on Python 3.9+, Flax 0.4.1+, PyTorch 2.0+, and TensorFlow 2.6+.
+In order to celebrate Transformers 100,000 stars, we wanted to put the spotlight on the
+community with the [awesome-transformers](./awesome-transformers.md) page which lists 100
+incredible projects built with Transformers.
 
-You should install 🤗 Transformers in a [virtual environment](https://docs.python.org/3/library/venv.html). If you're unfamiliar with Python virtual environments, check out the [user guide](https://packaging.python.org/guides/installing-using-pip-and-virtual-environments/).
+If you own or use a project that you believe should be part of the list, please open a PR to add it!
 
-First, create a virtual environment with the version of Python you're going to use and activate it.
+## Example models
 
-**macOS/Linux**
+You can test most of our models directly on their [Hub model pages](https://huggingface.co/models).
 
-```python -m venv env
-source env/bin/activate
-```
+Expand each modality below to see a few example models for various use cases.
 
-**Windows**
+<details>
+<summary>Audio</summary>
 
-``` python -m venv env
-env\Scripts\activate
-```
+- Audio classification with [Whisper](https://huggingface.co/openai/whisper-large-v3-turbo)
+- Automatic speech recognition with [Moonshine](https://huggingface.co/UsefulSensors/moonshine)
+- Keyword spotting with [Wav2Vec2](https://huggingface.co/superb/wav2vec2-base-superb-ks)
+- Speech to speech generation with [Moshi](https://huggingface.co/kyutai/moshiko-pytorch-bf16)
+- Text to audio with [MusicGen](https://huggingface.co/facebook/musicgen-large)
+- Text to speech with [Bark](https://huggingface.co/suno/bark)
 
-To use 🤗 Transformers, you must install at least one of Flax, PyTorch, or TensorFlow. Refer to the official installation guides for platform-specific commands:
+</details>
 
-[TensorFlow installation page](https://www.tensorflow.org/install/), 
-[PyTorch installation page](https://pytorch.org/get-started/locally/#start-locally) and/or [Flax](https://github.com/google/flax#quick-install) and [Jax](https://github.com/google/jax#installation) 
+<details>
+<summary>Computer vision</summary>
 
-When one of those backends has been installed, 🤗 Transformers can be installed using pip as follows:
+- Automatic mask generation with [SAM](https://huggingface.co/facebook/sam-vit-base)
+- Depth estimation with [DepthPro](https://huggingface.co/apple/DepthPro-hf)
+- Image classification with [DINO v2](https://huggingface.co/facebook/dinov2-base)
+- Keypoint detection with [SuperGlue](https://huggingface.co/magic-leap-community/superglue_outdoor)
+- Keypoint matching with [SuperGlue](https://huggingface.co/magic-leap-community/superglue)
+- Object detection with [RT-DETRv2](https://huggingface.co/PekingU/rtdetr_v2_r50vd)
+- Pose Estimation with [VitPose](https://huggingface.co/usyd-community/vitpose-base-simple)
+- Universal segmentation with [OneFormer](https://huggingface.co/shi-labs/oneformer_ade20k_swin_large)
+- Video classification with [VideoMAE](https://huggingface.co/MCG-NJU/videomae-large)
 
-```
-pip install transformers
-```
+</details>
 
-If you'd like to play with the examples or need the bleeding edge of the code and can't wait for a new release, you must [install the library from source](https://huggingface.co/docs/transformers/installation#installing-from-source).
+<details>
+<summary>Multimodal</summary>
 
-```
-git clone https://github.com/huggingface/transformers.git
-cd transformers
-pip install .
-```
+- Audio or text to text with [Qwen2-Audio](https://huggingface.co/Qwen/Qwen2-Audio-7B)
+- Document question answering with [LayoutLMv3](https://huggingface.co/microsoft/layoutlmv3-base)
+- Image or text to text with [Qwen-VL](https://huggingface.co/Qwen/Qwen2.5-VL-3B-Instruct)
+- Image captioning [BLIP-2](https://huggingface.co/Salesforce/blip2-opt-2.7b)
+- OCR-based document understanding with [GOT-OCR2](https://huggingface.co/stepfun-ai/GOT-OCR-2.0-hf)
+- Table question answering with [TAPAS](https://huggingface.co/google/tapas-base)
+- Unified multimodal understanding and generation with [Emu3](https://huggingface.co/BAAI/Emu3-Gen)
+- Vision to text with [Llava-OneVision](https://huggingface.co/llava-hf/llava-onevision-qwen2-0.5b-ov-hf)
+- Visual question answering with [Llava](https://huggingface.co/llava-hf/llava-1.5-7b-hf)
+- Visual referring expression segmentation with [Kosmos-2](https://huggingface.co/microsoft/kosmos-2-patch14-224)
 
-### With conda
+</details>
 
-🤗 Transformers can be installed using conda as follows:
+<details>
+<summary>NLP</summary>
 
-```shell script
-conda install conda-forge::transformers
-```
+- Masked word completion with [ModernBERT](https://huggingface.co/answerdotai/ModernBERT-base)
+- Named entity recognition with [Gemma](https://huggingface.co/google/gemma-2-2b)
+- Question answering with [Mixtral](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1)
+- Summarization with [BART](https://huggingface.co/facebook/bart-large-cnn)
+- Translation with [T5](https://huggingface.co/google-t5/t5-base)
+- Text generation with [Llama](https://huggingface.co/meta-llama/Llama-3.2-1B)
+- Text classification with [Qwen](https://huggingface.co/Qwen/Qwen2.5-0.5B)
 
-> **_NOTE:_** Installing `transformers` from the `huggingface` channel is deprecated.
-
-Follow the installation pages of Flax, PyTorch or TensorFlow to see how to install them with conda.
-
-> **_NOTE:_**  On Windows, you may be prompted to activate Developer Mode in order to benefit from caching. If this is not an option for you, please let us know in [this issue](https://github.com/huggingface/huggingface_hub/issues/1062).
-
-## Model architectures
-
-**[All the model checkpoints](https://huggingface.co/models)** provided by 🤗 Transformers are seamlessly integrated from the huggingface.co [model hub](https://huggingface.co/models), where they are uploaded directly by [users](https://huggingface.co/users) and [organizations](https://huggingface.co/organizations).
-
-Current number of checkpoints: ![](https://img.shields.io/endpoint?url=https://huggingface.co/api/shields/models&color=brightgreen)
-
-🤗 Transformers currently provides the following architectures: see [here](https://huggingface.co/docs/transformers/model_summary) for a high-level summary of each them.
-
-To check if each model has an implementation in Flax, PyTorch or TensorFlow, or has an associated tokenizer backed by the 🤗 Tokenizers library, refer to [this table](https://huggingface.co/docs/transformers/index#supported-frameworks).
-
-These implementations have been tested on several datasets (see the example scripts) and should match the performance of the original implementations. You can find more details on performance in the Examples section of the [documentation](https://github.com/huggingface/transformers/tree/main/examples).
-
-
-## Learn more
-
-| Section | Description |
-|-|-|
-| [Documentation](https://huggingface.co/docs/transformers/) | Full API documentation and tutorials |
-| [Task summary](https://huggingface.co/docs/transformers/task_summary) | Tasks supported by 🤗 Transformers |
-| [Preprocessing tutorial](https://huggingface.co/docs/transformers/preprocessing) | Using the `Tokenizer` class to prepare data for the models |
-| [Training and fine-tuning](https://huggingface.co/docs/transformers/training) | Using the models provided by 🤗 Transformers in a PyTorch/TensorFlow training loop and the `Trainer` API |
-| [Quick tour: Fine-tuning/usage scripts](https://github.com/huggingface/transformers/tree/main/examples) | Example scripts for fine-tuning models on a wide range of tasks |
-| [Model sharing and uploading](https://huggingface.co/docs/transformers/model_sharing) | Upload and share your fine-tuned models with the community |
+</details>
 
 ## Citation
 

SECURITY.md

@@ -27,13 +27,6 @@ These models require the `trust_remote_code=True` parameter to be set when using
 the content of the modeling files when using this argument. We recommend setting a revision in order to ensure you
 protect yourself from updates on the repository.
 
-#### Tools
-
-Through the `Agent` framework, remote tools can be downloaded to be used by the Agent. You're to specify these tools
-yourself, but please keep in mind that their code will be run on your machine if the Agent chooses to run them.
-
-Please inspect the code of the tools before passing them to the Agent to protect your runtime and local setup.
-
 ## Reporting a Vulnerability
 
 Feel free to submit vulnerability reports to [security@huggingface.co](mailto:security@huggingface.co), where someone from the HF security team will review and recommend next steps. If reporting a vulnerability specific to open source, please note [Huntr](https://huntr.com) is a vulnerability disclosure program for open source software.

benchmark/README.md

@@ -12,7 +12,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
 ## Writing metrics to the database
 
-`MetricRecorder` is thread-safe, in the sense of the python [`Thread`](https://docs.python.org/3/library/threading.html#threading.Thread). This means you can start a background thread to do the readings on the device measurements while not blocking the main thread to execute the model measurements.
+`MetricsRecorder` is thread-safe, in the sense of the python [`Thread`](https://docs.python.org/3/library/threading.html#threading.Thread). This means you can start a background thread to do the readings on the device measurements while not blocking the main thread to execute the model measurements.
 
 cf [`llama.py`](./llama.py) to see an example of this in practice.
 

benchmark/benchmark.py

@@ -90,7 +90,7 @@ def summarize(run_dir, metrics, expand_metrics=False):
 
         model = benchmark.config.backend["model"]
 
-        # Ths looks like `benchmark.input_shapes.batch_size=1,benchmark.input_shapes.sequence_length=5`.
+        # This looks like `benchmark.input_shapes.batch_size=1,benchmark.input_shapes.sequence_length=5`.
         # (we rely on the usage of hydra's `${hydra.job.override_dirname}`.)
         benchmark_name = re.sub(f"backend.model={model},*", "", report_dir)
         benchmark_name = str(Path(benchmark_name).parts[-1])

benchmark/benchmarks_entrypoint.py

@@ -3,7 +3,6 @@ import importlib.util
 import logging
 import os
 from typing import Dict
-import psycopg2
 import sys
 
 from psycopg2.extras import Json

benchmark/llama.py

@@ -118,7 +118,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
         with torch.no_grad():
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + num_tokens_to_generate,
@@ -144,7 +144,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + num_tokens_to_generate,
@@ -187,7 +187,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             # TODO use  decode_one_token(model, input_id.clone(), cache_position) for verification
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + num_tokens_to_generate + 10,
@@ -204,7 +204,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             time_to_first_token = end - start
             logger.info(f"completed first compile generation in: {time_to_first_token}s")
             cache_position += 1
-            all_generated_tokens += next_token.clone().detach().cpu().tolist()
+            all_generated_tokens += next_token.tolist()
 
             cache_position = torch.tensor([seq_length], device=device)
             ### First compile, decoding
@@ -215,9 +215,9 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             torch.cuda.synchronize()
             end = perf_counter()
             time_to_second_token = end - start
-            logger.info(f"completed second compile generation in: {time_to_first_token}s")
+            logger.info(f"completed second compile generation in: {time_to_second_token}s")
             cache_position += 1
-            all_generated_tokens += next_token.clone().detach().cpu().tolist()
+            all_generated_tokens += next_token.tolist()
 
             ### Second compile, decoding
             start = perf_counter()
@@ -227,15 +227,15 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             torch.cuda.synchronize()
             end = perf_counter()
             time_to_third_token = end - start
-            logger.info(f"completed third compile forward in: {time_to_first_token}s")
+            logger.info(f"completed third compile forward in: {time_to_third_token}s")
             cache_position += 1
-            all_generated_tokens += next_token.clone().detach().cpu().tolist()
+            all_generated_tokens += next_token.tolist()
 
             ### Using cuda graphs decoding
 
             start = perf_counter()
             for _ in range(1, num_tokens_to_generate):
-                all_generated_tokens += next_token.clone().detach().cpu().tolist()
+                all_generated_tokens += next_token.tolist()
                 next_token = decode_one_token(
                     model, next_token.clone(), cache_position=cache_position, past_key_values=past_key_values
                 )
@@ -254,7 +254,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + 128,
@@ -271,7 +271,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + 128,
@@ -287,23 +287,23 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
 
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + 128,
             )
 
-            # 3nd call
+            # 3rd call
             start = perf_counter()
             output = model.generate(**inputs, past_key_values=past_key_values)
             end = perf_counter()
             third_compile_generate_time = end - start
-            logger.info(f"completed second compile generation in: {third_compile_generate_time}s")
+            logger.info(f"completed third compile generation in: {third_compile_generate_time}s")
             logger.info(f"generated: {tokenizer.batch_decode(output.cpu().tolist())}")
 
             past_key_values = StaticCache(
                 model.config,
-                batch_size=batch_size,
+                max_batch_size=batch_size,
                 device=device,
                 dtype=torch.float16,
                 max_cache_len=seq_length + 128,
@@ -313,7 +313,7 @@ def run_benchmark(logger: Logger, branch: str, commit_id: str, commit_msg: str,
             output = model.generate(**inputs, past_key_values=past_key_values)
             end = perf_counter()
             fourth_compile_generate_time = end - start
-            logger.info(f"completed second compile generation in: {fourth_compile_generate_time}s")
+            logger.info(f"completed fourth compile generation in: {fourth_compile_generate_time}s")
             logger.info(f"generated: {tokenizer.batch_decode(output.cpu().tolist())}")
 
         metrics_recorder.collect_model_measurements(

conftest.py

@@ -46,10 +46,6 @@ NOT_DEVICE_TESTS = {
     "test_keep_in_fp32_modules",
     "test_gradient_checkpointing_backward_compatibility",
     "test_gradient_checkpointing_enable_disable",
-    "test_save_load_fast_init_from_base",
-    "test_fast_init_context_manager",
-    "test_fast_init_tied_embeddings",
-    "test_save_load_fast_init_to_base",
     "test_torch_save_load",
     "test_initialization",
     "test_forward_signature",
@@ -70,7 +66,6 @@ NOT_DEVICE_TESTS = {
     "ModelTester::test_pipeline_",
     "/repo_utils/",
     "/utils/",
-    "/agents/",
 }
 
 # allow having multiple repository checkouts and not needing to remember to rerun
@@ -87,7 +82,6 @@ def pytest_configure(config):
     config.addinivalue_line("markers", "is_pipeline_test: mark test to run only when pipelines are tested")
     config.addinivalue_line("markers", "is_staging_test: mark test to run only in the staging environment")
     config.addinivalue_line("markers", "accelerate_tests: mark test that require accelerate")
-    config.addinivalue_line("markers", "agent_tests: mark the agent tests that are run on their specific schedule")
     config.addinivalue_line("markers", "not_device_test: mark the tests always running on cpu")
 
 

docker/README.md

@@ -2,8 +2,8 @@
 
 In this folder you will find various docker files, and some subfolders. 
 - dockerfiles (ex: `consistency.dockerfile`) present under `~/docker` are used for our "fast" CIs. You should be able to use them for tasks that only need CPU. For example `torch-light` is a very light weights container (703MiB). 
-- subfloder contain dockerfiles used for our `slow` CIs, which *can* be used for GPU tasks, but they are **BIG** as they were not specifically designed for a single model / single task. Thus the `~/docker/transformers-pytorch-gpu` includes additional dependencies to allow us to run ALL model tests (say `librosa` or `tesseract`, which you do not need to run LLMs)
+- subfolders contain dockerfiles used for our `slow` CIs, which *can* be used for GPU tasks, but they are **BIG** as they were not specifically designed for a single model / single task. Thus the `~/docker/transformers-pytorch-gpu` includes additional dependencies to allow us to run ALL model tests (say `librosa` or `tesseract`, which you do not need to run LLMs)
 
 Note that in both case, you need to run `uv pip install -e .`, which should take around 5 seconds. We do it outside the dockerfile for the need of our CI: we checkout a new branch each time, and the `transformers` code is thus updated. 
 
-We are open to contribution, and invite the community to create dockerfiles with potential arguments that properly choose extras depending on the model's dependencies! :hugs: 
+We are open to contribution, and invite the community to create dockerfiles with potential arguments that properly choose extras depending on the model's dependencies! :hugs: 

docker/consistency.dockerfile

@@ -5,12 +5,12 @@ ARG REF=main
 RUN apt-get update && apt-get install -y time git g++ pkg-config make git-lfs
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools GitPython
-RUN pip install --no-cache-dir --upgrade 'torch' 'torchaudio' 'torchvision' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir --upgrade 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
 # tensorflow pin matching setup.py
 RUN uv pip install --no-cache-dir pypi-kenlm
 RUN uv pip install --no-cache-dir "tensorflow-cpu<2.16" "tf-keras<2.16"
 RUN uv pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,quality,testing,torch-speech,vision]"
 RUN git lfs install
 
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean
+RUN uv pip uninstall transformers
+RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/custom-tokenizers.dockerfile

@@ -1,5 +1,6 @@
 FROM python:3.9-slim
 ENV PYTHONDONTWRITEBYTECODE=1
+ARG REF=main
 USER root
 RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git cmake wget xz-utils build-essential g++5 libprotobuf-dev protobuf-compiler
 ENV UV_PYTHON=/usr/local/bin/python
@@ -15,12 +16,12 @@ RUN cmake .. -DCMAKE_INSTALL_PREFIX=/usr/local
 RUN make install -j 10
 
 
-RUN uv pip install --no-cache --upgrade 'torch' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-cache-dir  --no-deps accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install  --no-cache-dir "transformers[ja,testing,sentencepiece,jieba,spacy,ftfy,rjieba]" unidic unidic-lite
+RUN uv pip install --no-cache --upgrade 'torch==2.6.0' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir  --no-deps accelerate --extra-index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install  --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[ja,testing,sentencepiece,jieba,spacy,ftfy,rjieba]" unidic unidic-lite
 # spacy is not used so not tested. Causes to failures. TODO fix later
 RUN python3 -m unidic download
-RUN pip uninstall -y transformers
+RUN uv pip uninstall transformers
 
 RUN apt-get clean && rm -rf /var/lib/apt/lists/*
-RUN apt remove -y g++ cmake  xz-utils libprotobuf-dev protobuf-compiler
+RUN apt remove -y g++ cmake  xz-utils libprotobuf-dev protobuf-compiler

docker/examples-tf.dockerfile

@@ -1,12 +1,13 @@
 FROM python:3.9-slim
 ENV PYTHONDONTWRITEBYTECODE=1
+ARG REF=main
 USER root
 RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git
 RUN apt-get install -y g++ cmake
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv
 RUN uv pip install --no-cache-dir -U pip setuptools albumentations seqeval
-RUN pip install  --upgrade --no-cache-dir "transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
-RUN uv pip install --no-cache-dir  "protobuf==3.20.3" 
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*
+RUN uv pip install  --upgrade --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
+RUN uv pip install --no-cache-dir  "protobuf==3.20.3"
+RUN uv pip uninstall transformers
+RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/examples-torch.dockerfile

@@ -1,11 +1,12 @@
 FROM python:3.9-slim
 ENV PYTHONDONTWRITEBYTECODE=1
+ARG REF=main
 USER root
 RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN uv pip install --no-cache-dir librosa "transformers[sklearn,sentencepiece,vision,testing]" seqeval albumentations jiwer
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*
+RUN uv pip install --no-cache-dir 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing]" seqeval albumentations jiwer
+RUN uv pip uninstall transformers
+RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/exotic-models.dockerfile

@@ -5,13 +5,13 @@ USER root
 RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git libgl1-mesa-glx libgl1 g++ tesseract-ocr
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv &&  uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
 RUN uv pip install --no-cache-dir  --no-deps timm accelerate
 RUN pip install -U --upgrade-strategy eager --no-cache-dir pytesseract python-Levenshtein opencv-python nltk
 # RUN uv pip install --no-cache-dir natten==0.15.1+torch210cpu -f https://shi-labs.com/natten/wheels
-RUN pip install  --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[testing, vision]" 'scikit-learn' 'torch-stft' 'nose'  'dataset'
+RUN uv pip install  --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[testing, vision]" 'scikit-learn' 'torch-stft' 'nose'  'dataset'
 # RUN git clone https://github.com/facebookresearch/detectron2.git
 # RUN python3 -m pip install --no-cache-dir -e detectron2
-RUN pip install 'git+https://github.com/facebookresearch/detectron2.git@92ae9f0b92aba5867824b4f12aa06a22a60a45d3'
-RUN pip uninstall -y transformers
+RUN uv pip install 'git+https://github.com/facebookresearch/detectron2.git@92ae9f0b92aba5867824b4f12aa06a22a60a45d3' --no-build-isolation
+RUN uv pip uninstall transformers
 RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/jax-light.dockerfile

@@ -5,6 +5,6 @@ USER root
 RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git g++ cmake
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv &&  uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,testing,sentencepiece,flax-speech,vision]"
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean
+RUN uv pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,testing,sentencepiece,flax-speech,vision]"
+RUN uv pip uninstall transformers
+RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/pipeline-tf.dockerfile

@@ -5,6 +5,6 @@ USER root
 RUN apt-get update && apt-get install -y libsndfile1-dev espeak-ng time git cmake g++
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]"
+RUN uv pip install --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,tf-cpu,testing,sentencepiece,tf-speech,vision]"
 RUN uv pip install --no-cache-dir  "protobuf==3.20.3" tensorflow_probability
-RUN apt-get clean && rm -rf /var/lib/apt/lists/*
+RUN apt-get clean && rm -rf /var/lib/apt/lists/*

docker/pipeline-torch.dockerfile

@@ -5,7 +5,7 @@ USER root
 RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git pkg-config openssh-client git
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
+RUN uv pip install --no-cache-dir --upgrade 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu
 RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing]"
-RUN pip uninstall -y transformers
+RUN uv pip uninstall transformers

docker/quality.dockerfile

@@ -6,4 +6,4 @@ RUN apt-get update && apt-get install -y time git
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip install uv &&  uv venv
 RUN uv pip install --no-cache-dir -U pip setuptools GitPython "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[ruff]" urllib3
-RUN apt-get install -y jq curl && apt-get clean && rm -rf /var/lib/apt/lists/*
+RUN apt-get install -y jq curl && apt-get clean && rm -rf /var/lib/apt/lists/*

docker/tf-light.dockerfile

@@ -6,7 +6,7 @@ RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-de
 RUN apt-get install -y  cmake
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install  --upgrade --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
+RUN uv pip install  --upgrade --no-cache-dir "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,testing,sentencepiece,tf-speech,vision]"
 RUN uv pip install --no-cache-dir  "protobuf==3.20.3"
-RUN pip uninstall -y transformers
+RUN uv pip uninstall transformers
 RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/torch-jax-light.dockerfile

@@ -6,11 +6,11 @@ RUN apt-get update &&  apt-get install -y libsndfile1-dev espeak-ng time git g++
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
 RUN uv pip install --no-deps accelerate
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
-RUN pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,audio,sklearn,sentencepiece,vision,testing]"
+RUN uv pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir "scipy<1.13" "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[flax,audio,sklearn,sentencepiece,vision,testing]"
 
 
 # RUN pip install --no-cache-dir "scipy<1.13" "transformers[flax,testing,sentencepiece,flax-speech,vision]"
 
-RUN pip uninstall -y transformers
+RUN uv pip uninstall transformers
 RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/torch-light.dockerfile

@@ -5,7 +5,7 @@ USER root
 RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-dev espeak-ng time git g++ cmake pkg-config openssh-client git git-lfs
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir --upgrade 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
 RUN uv pip install --no-deps timm accelerate --extra-index-url https://download.pytorch.org/whl/cpu
-RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing,tiktoken,num2words]"
-RUN pip uninstall -y transformers
+RUN uv pip install --no-cache-dir librosa "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[sklearn,sentencepiece,vision,testing,tiktoken,num2words,video]"
+RUN uv pip uninstall transformers

docker/torch-tf-light.dockerfile

@@ -7,13 +7,13 @@ RUN apt-get update &&  apt-get install -y --no-install-recommends libsndfile1-de
 ENV UV_PYTHON=/usr/local/bin/python
 RUN pip --no-cache-dir install uv && uv venv && uv pip install --no-cache-dir -U pip setuptools
 RUN uv pip install --no-cache-dir  --no-deps accelerate --extra-index-url https://download.pytorch.org/whl/cpu 
-RUN pip install --no-cache-dir 'torch' 'torchvision' 'torchaudio' --index-url https://download.pytorch.org/whl/cpu
+RUN uv pip install --no-cache-dir 'torch==2.6.0' 'torchaudio==2.6.0' 'torchvision==0.21.0' --index-url https://download.pytorch.org/whl/cpu
 RUN git lfs install
 
 RUN uv pip install --no-cache-dir pypi-kenlm
-RUN pip install --no-cache-dir  "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,sentencepiece,vision,testing]"
+RUN uv pip install --no-cache-dir  "git+https://github.com/huggingface/transformers.git@${REF}#egg=transformers[tf-cpu,sklearn,sentencepiece,vision,testing]"
 RUN uv pip install --no-cache-dir  "protobuf==3.20.3" librosa
 
 
-RUN pip uninstall -y transformers
-RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean
+RUN uv pip uninstall transformers
+RUN apt-get clean && rm -rf /var/lib/apt/lists/* && apt-get autoremove && apt-get autoclean

docker/transformers-all-latest-gpu/Dockerfile

@@ -14,6 +14,8 @@ ARG PYTORCH='2.6.0'
 ARG INTEL_TORCH_EXT='2.3.0'
 # Example: `cu102`, `cu113`, etc.
 ARG CUDA='cu121'
+# Disable kernel mapping for now until all tests pass
+ENV DISABLE_KERNEL_MAPPING=1
 
 RUN apt update
 RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg git-lfs
@@ -57,7 +59,8 @@ RUN python3 -m pip uninstall -y ninja
 
 # For `dinat` model
 # The `XXX` part in `torchXXX` needs to match `PYTORCH` (to some extent)
-RUN python3 -m pip install --no-cache-dir natten==0.15.1+torch220$CUDA -f https://shi-labs.com/natten/wheels
+# pin `0.17.4` otherwise `cannot import name 'natten2dav' from 'natten.functional'`
+RUN python3 -m pip install --no-cache-dir natten==0.17.4+torch250cu121 -f https://shi-labs.com/natten/wheels
 
 # For `nougat` tokenizer
 RUN python3 -m pip install --no-cache-dir python-Levenshtein

docker/transformers-pytorch-deepspeed-latest-gpu/Dockerfile

@@ -1,12 +1,12 @@
-# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.11-py3
+# https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-08.html
+FROM nvcr.io/nvidia/pytorch:24.08-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
-ARG PYTORCH='2.2.0'
+ARG PYTORCH='2.6.0'
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu126'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -15,7 +15,8 @@ RUN python3 -m pip install --no-cache-dir --upgrade pip
 ARG REF=main
 RUN git clone https://github.com/huggingface/transformers && cd transformers && git checkout $REF
 
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
+# `datasets` requires pandas, pandas has some modules compiled with numpy=1.x causing errors
+RUN python3 -m pip install --no-cache-dir './transformers[deepspeed-testing]' 'pandas<2' 'numpy<2'
 
 # Install latest release PyTorch
 # (PyTorch must be installed before pre-compiling any DeepSpeed c++/cuda ops.)

docker/transformers-pytorch-deepspeed-nightly-gpu/Dockerfile

@@ -1,11 +1,11 @@
 # https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-23-11.html#rel-23-11
-FROM nvcr.io/nvidia/pytorch:23.11-py3
+FROM nvcr.io/nvidia/pytorch:24.08-py3
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
 
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu121'
+ARG CUDA='cu126'
 
 RUN apt -y update
 RUN apt install -y libaio-dev
@@ -21,7 +21,8 @@ RUN python3 -m pip uninstall -y torch torchvision torchaudio
 # (https://www.deepspeed.ai/tutorials/advanced-install/#pre-install-deepspeed-ops)
 RUN python3 -m pip install --no-cache-dir -U --pre torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/nightly/$CUDA
 
-RUN python3 -m pip install --no-cache-dir ./transformers[deepspeed-testing]
+# `datasets` requires pandas, pandas has some modules compiled with numpy=1.x causing errors
+RUN python3 -m pip install --no-cache-dir './transformers[deepspeed-testing]' 'pandas<2' 'numpy<2'
 
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
 

docker/transformers-quantization-latest-gpu/Dockerfile

@@ -1,4 +1,4 @@
-FROM nvidia/cuda:11.8.0-cudnn8-devel-ubuntu22.04
+FROM nvidia/cuda:12.1.1-cudnn8-devel-ubuntu22.04
 LABEL maintainer="Hugging Face"
 
 ARG DEBIAN_FRONTEND=noninteractive
@@ -9,9 +9,11 @@ SHELL ["sh", "-lc"]
 # The following `ARG` are mainly used to specify the versions explicitly & directly in this docker file, and not meant
 # to be used as arguments for docker build (so far).
 
-ARG PYTORCH='2.5.1'
+ARG PYTORCH='2.6.0'
 # Example: `cu102`, `cu113`, etc.
-ARG CUDA='cu118'
+ARG CUDA='cu121'
+# Disable kernel mapping for quantization tests
+ENV DISABLE_KERNEL_MAPPING=1
 
 RUN apt update
 RUN apt install -y git libsndfile1-dev tesseract-ocr espeak-ng python3 python3-pip ffmpeg
@@ -26,8 +28,6 @@ RUN echo torch=$VERSION
 # Currently, let's just use their latest releases (when `torch` is installed with a release version)
 RUN python3 -m pip install --no-cache-dir -U $VERSION torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/$CUDA
 
-RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch]
-
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/accelerate@main#egg=accelerate
 
 # needed in bnb and awq
@@ -36,10 +36,9 @@ RUN python3 -m pip install --no-cache-dir einops
 # Add bitsandbytes for mixed int8 testing
 RUN python3 -m pip install --no-cache-dir bitsandbytes
 
-# Add auto-gptq for gtpq quantization testing, installed from source for pytorch==2.5.1 compatibility
-# TORCH_CUDA_ARCH_LIST="7.5+PTX" is added to make the package compile for Tesla T4 gpus available for the CI.
-RUN pip install gekko
-RUN git clone https://github.com/PanQiWei/AutoGPTQ.git && cd AutoGPTQ && TORCH_CUDA_ARCH_LIST="7.5+PTX" python3 setup.py install
+# Add gptqmodel for gtpq quantization testing, installed from source for pytorch==2.6.0 compatibility
+RUN python3 -m pip install lm_eval
+RUN git clone https://github.com/ModelCloud/GPTQModel.git && cd GPTQModel && pip install -v . --no-build-isolation
 
 # Add optimum for gptq quantization testing
 RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/optimum@main#egg=optimum
@@ -51,10 +50,11 @@ RUN python3 -m pip install --no-cache-dir git+https://github.com/huggingface/pef
 RUN python3 -m pip install --no-cache-dir aqlm[gpu]==1.0.2
 
 # Add vptq for quantization testing
-RUN python3 -m pip install --no-cache-dir vptq
+RUN pip install vptq
 
 # Add spqr for quantization testing
-RUN python3 -m pip install --no-cache-dir spqr_quant[gpu]
+# Commented for now as No matching distribution found we need to reach out to the authors
+# RUN python3 -m pip install --no-cache-dir spqr_quant[gpu]
 
 # Add hqq for quantization testing
 RUN python3 -m pip install --no-cache-dir hqq
@@ -63,22 +63,33 @@ RUN python3 -m pip install --no-cache-dir hqq
 RUN python3 -m pip install --no-cache-dir gguf
 
 # Add autoawq for quantization testing
-# >=v0.2.7 needed for compatibility with transformers > 4.46
-RUN python3 -m pip install --no-cache-dir https://github.com/casper-hansen/AutoAWQ/releases/download/v0.2.7.post2/autoawq-0.2.7.post2-py3-none-any.whl
+# New release v0.2.8
+RUN python3 -m pip install --no-cache-dir autoawq[kernels]
 
 # Add quanto for quantization testing
 RUN python3 -m pip install --no-cache-dir optimum-quanto
 
 # Add eetq for quantization testing
-RUN python3 -m pip install git+https://github.com/NetEase-FuXi/EETQ.git
+RUN git clone https://github.com/NetEase-FuXi/EETQ.git && cd EETQ/ && git submodule update --init --recursive && pip install .
 
-# Add flute-kernel and fast_hadamard_transform for quantization testing
-RUN python3 -m pip install --no-cache-dir flute-kernel==0.3.0 -i https://flute-ai.github.io/whl/cu118
-RUN python3 -m pip install --no-cache-dir fast_hadamard_transform==1.0.4.post1
+# # Add flute-kernel and fast_hadamard_transform for quantization testing
+# # Commented for now as they cause issues with the build
+# # TODO: create a new workflow to test them
+# RUN python3 -m pip install --no-cache-dir flute-kernel==0.4.1
+# RUN python3 -m pip install --no-cache-dir git+https://github.com/Dao-AILab/fast-hadamard-transform.git
 
 # Add compressed-tensors for quantization testing
 RUN python3 -m pip install --no-cache-dir compressed-tensors
 
+# Add AMD Quark for quantization testing
+RUN python3 -m pip install --no-cache-dir amd-quark
+
+# Add AutoRound for quantization testing
+RUN python3 -m pip install --no-cache-dir "auto-round>=0.5.0"
+
+# Add transformers in editable mode
+RUN python3 -m pip install --no-cache-dir -e ./transformers[dev-torch]
+
 # When installing in editable mode, `transformers` is not recognized as a package.
 # this line must be added in order for python to be aware of transformers.
 RUN cd transformers && python3 setup.py develop

docs/source/ar/_toctree.yml

@@ -23,8 +23,6 @@
     title: تحميل النماذج المخصصة وتدريبها باستخدام 🤗 PEFT
   - local: model_sharing
     title: مشاركة نموذجك
-  - local: agents
-    title: الوكلاء
   - local: llm_tutorial
     title: التوليد باستخدام LLMs
   - local: conversations
@@ -252,8 +250,6 @@
   title: أطر مفاهيمية
 # - sections:
 #   - sections:
-#     - local: main_classes/agent
-#       title: الوكلاء والأدوات
 #     - local: model_doc/auto
 #       title: فئات يتم إنشاؤها ديناميكيًا
 #     - local: main_classes/backbones

docs/source/ar/agents.md

@@ -1,539 +0,0 @@
-# الوكلاء والأدوات
-
-[[open-in-colab]]
-
-### ما هو الوكيل؟
-
-يمكن للنظم اللغوية الكبيرة (LLMs) التي تم تدريبها على أداء [نمذجة اللغة السببية](./tasks/language_modeling.) التعامل مع مجموعة واسعة من المهام، ولكنها غالبًا ما تواجه صعوبات في المهام الأساسية مثل المنطق والحساب والبحث. وعندما يتم استدعاؤها في مجالات لا تؤدي فيها أداءً جيدًا، فإنها غالبًا ما تفشل في توليد الإجابة التي نتوقعها منها.
-
-يتمثل أحد النهج للتغلب على هذا القصور في إنشاء "وكيل".
-
-الوكيل هو نظام يستخدم LLM كمحرك له، ولديه حق الوصول إلى وظائف تسمى "أدوات".
-
-هذه "الأدوات" هي وظائف لأداء مهمة، وتحتوي على جميع الأوصاف اللازمة للوكيل لاستخدامها بشكل صحيح.
-
-يمكن برمجة الوكيل للقيام بما يلي:
-- وضع سلسلة من الإجراءات/الأدوات وتشغيلها جميعًا في نفس الوقت مثل [`CodeAgent`] على سبيل المثال
-- التخطيط للاجراءات/الأدوات وتنفيذها واحدة تلو الأخرى والانتظار حتى انتهاء كل إجراء قبل إطلاق التالي مثل [`ReactJsonAgent`] على سبيل المثال
-
-### أنواع الوكلاء
-
-#### الوكيل البرمجي (Code agent)
-
-يتمتع هذا الوكيل يتبع خطوات محددة: أولًا، يخطط لسلسلة من الإجراءات التي يريد تنفيذها، ثم شفرة Python لتنفيذ جميع الإجراءات في نفس الوقت. وهو يتعامل بشكل أصلي مع أنواع مختلفة من المدخلات والمخرجات للأدوات التي يستخدمها، وبالتالي فهو الخيار الموصى به للمهام متعددة الوسائط.
-
-#### وكلاء التفاعل
-
-هذا هو الوكيل الذي يتم اللجوء إليه لحل مهام الاستدلال، حيث يجعل إطار ReAct ([Yao et al.، 2022](https://huggingface.co/papers/2210.03629)) من الكفاءة حقًا التفكير على أساس ملاحظاته السابقة.
-
-نقوم بتنفيذ إصدارين من ReactJsonAgent: 
-- [`ReactJsonAgent`] يقوم بتوليد استدعاءات الأدوات كـ JSON في إخراجها.
-- [`ReactCodeAgent`] هو نوع جديد من ReactJsonAgent يقوم بتوليد استدعاءات أدواته كمقاطع من التعليمات البرمجية، والتي تعمل بشكل جيد حقًا مع LLMs التي تتمتع بأداء  قوي في البرمجة.
-
-> [!TIP]
-> اقرأ منشور المدونة [Open-source LLMs as LangChain Agents](https://huggingface.co/blog/open-source-llms-as-agents) لمعرفة المزيد عن وكيل ReAct.
-
-![إطار عمل وكيل ReAct](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/open-source-llms-as-agents/ReAct.png)
-
-على سبيل المثال، إليك كيف يعمل وكيل ReAct Code طريقه من خلال السؤال التالي.
-
-```py3
->>> agent.run(
-...     "How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?",
-... )
-=====New task=====
-How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?
-====Agent is executing the code below:
-bert_blocks = search(query="number of blocks in BERT base encoder")
-print("BERT blocks:", bert_blocks)
-====
-Print outputs:
-BERT blocks: twelve encoder blocks
-
-====Agent is executing the code below:
-attention_layer = search(query="number of layers in Attention is All You Need")
-print("Attention layers:", attention_layer)
-====
-Print outputs:
-Attention layers: Encoder: The encoder is composed of a stack of N = 6 identical layers. Each layer has two sub-layers. The first is a multi-head self-attention mechanism, and the second is a simple, position- 2 Page 3 Figure 1: The Transformer - model architecture.
-
-====Agent is executing the code below:
-bert_blocks = 12
-attention_layers = 6
-diff = bert_blocks - attention_layers
-print("Difference in blocks:", diff)
-final_answer(diff)
-====
-
-Print outputs:
-Difference in blocks: 6
-
-Final answer: 6
-```
-
-### كيف يمكنني بناء وكيل؟
-
-لتهيئة وكيل، تحتاج إلى هذه الوسائط:
-
-- نموذج لغوي كبير (LLM) يشكل المحرك الأساسي للوكيل. الوكيل نفسه ليس النموذج اللغوي، بل هو برنامج يستخدم النموذج اللغوي كمحرك له.
-- موجه النظام (system prompt): هذه هي التعليمات التي يتم إعطاؤها للنموذج اللغوي لإنشاء مخرجاته.
-- صندوق أدوات (toolbox) يختار الوكيل منه الأدوات لتنفيذها
-- محلل (parser) لاستخراج الأدوات التي يجب استدعاؤها من مخرجات النموذج اللغوي LLM والأدوات التي يجب استخدامها
-
-عند تهيئة نظام الوكيل، يتم استخدام سمات الأداة لإنشاء وصف للأداة، ثم يتم دمجها في موجه النظام الخاص `system_prompt` للوكيل لإعلامه بالأدوات التي يمكنه استخدامها ولماذا.
-
-للبدء، يرجى تثبيت `agents` الإضافية لتثبيت جميع التبعيات الافتراضية.
-
-```bash
-pip install transformers[agents]
-```
-
-قم ببناء محرك LLM الخاص بك من خلال تعريف طريقة `llm_engine` التي تقبل قائمة من [الرسائل](./chat_templating.) وتعيد النص. يجب أن تقبل هذه الدالة القابلة للاستدعاء أيضًا معامل `stop` يشير إلى متى يجب التوقف عن التوليد.
-
-```python
-from huggingface_hub import login, InferenceClient
-
-login("<YOUR_HUGGINGFACEHUB_API_TOKEN>")
-
-client = InferenceClient(model="meta-llama/Meta-Llama-3-70B-Instruct")
-
-def llm_engine(messages, stop_sequences=["Task"]) -> str:
-    response = client.chat_completion(messages, stop=stop_sequences, max_tokens=1000)
-    answer = response.choices[0].message.content
-    return answer
-```
-
-يمكنك استخدام أي طريقة `llm_engine` طالما أنها:
-1. يتبع تنسيق [رسائل](./chat_templating.md) لإدخاله (`List [Dict [str، str]]`) ويعيد `str`
-2. يتوقف عن توليد المخراجات من التسلسلات التي تم تمريرها في معامل `stop`
-
-أنت بحاجة أيضًا إلى معامل "الأدوات" الذي يقبل قائمة من "الأدوات". يمكنك توفير قائمة فارغة لـ "الأدوات"، ولكن استخدم صندوق الأدوات الافتراضي مع معامل اختياري `add_base_tools=True`.
-
-الآن يمكنك إنشاء وكيل، مثل [`CodeAgent`], وتشغيله. ولتسهيل الأمر، نقدم أيضًا فئة [`HfEngine`] التي تستخدم `huggingface_hub.InferenceClient` بشكل مخفى.
-
-```python
-from transformers import CodeAgent, HfEngine
-
-llm_engine = HfEngine(model="meta-llama/Meta-Llama-3-70B-Instruct")
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and return the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-هذه الميزة ستكون مفيدة في حالة الحاجة الملحة! يمكنك حتى ترك معامل `llm_engine` غير محدد، وسيتم إنشاء [`HfEngine`] بشكل تلقائي.
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], add_base_tools=True)
-
-agent.run(
-    "Could you translate this sentence from French, say it out loud and give me the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-لاحظ أننا استخدمنا معامل "sentence" إضافي: يمكنك تمرير النص كمعامل إضافي إلى النموذج.
-
-يمكنك أيضًا استخدام هذا للإشارة إلى مسار الملفات المحلية أو البعيدة للنموذج لاستخدامها:
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-
-agent.run("Why does Mike not know many people in New York?", audio="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/recording.mp3")
-```
-
-
-تم تحديد موجه النظام ومحلل المخرجات تلقائيًا، ولكن يمكنك فحصهما بسهولة عن طريق استدعاء `system_prompt_template` على وكيلك.
-
-```python
-print(agent.system_prompt_template)
-```
-
-من المهم أن تشرح بأكبر قدر ممكن من الوضوح المهمة التي تريد تنفيذها.
-كل عملية [`~Agent.run`] مستقلة، وبما أن الوكيل مدعوم من LLM، فقد تؤدي الاختلافات الطفيفة في موجهك إلى نتائج مختلفة تمامًا.
-يمكنك أيضًا تشغيل وكيل بشكل متتالي لمهام مختلفة: في كل مرة يتم فيها إعادة تهيئة سمتي `agent.task` و`agent.logs`.
-
-
-#### تنفيذ التعليمات البرمجية
-
-يقوم مفسر Python بتنفيذ التعليمات البرمجية على مجموعة من المدخلات التي يتم تمريرها جنبًا إلى جنب مع أدواتك.
-يجب أن يكون هذا الأمر آمنًا لأن الوظائف الوحيدة التي يمكن استدعاؤها هي الأدوات التي قدمتها (خاصة إذا كانت أدوات من Hugging Face فقط) ووظيفة الطباعة، لذا فأنت مقيد بالفعل بما يمكن تنفيذه.
-
-مفسر Python لا يسمح أيضًا باستدعاء دوال بشكل افتراضي خارج قائمة آمنة، لذا فإن جميع الهجمات الأكثر وضوحًا لا ينبغي أن تكون مشكلة.
-يمكنك أيضًا الإذن باستيرادات إضافية عن طريق تمرير الوحدات النمطية المصرح بها كقائمة من السلاسل في معامل  `additional_authorized_imports` عند تهيئة [`ReactCodeAgent`] أو [`CodeAgent`]:
-
-```py
->>> from transformers import ReactCodeAgent
-
->>> agent = ReactCodeAgent(tools=[], additional_authorized_imports=['requests', 'bs4'])
->>> agent.run("Could you get me the title of the page at url 'https://huggingface.co/blog'?")
-
-(...)
-'Hugging Face – Blog'
-```
-
-سيتم إيقاف التنفيذ عند أي رمز يحاول تنفيذ عملية غير قانونية أو إذا كان هناك خطأ Python عادي في التعليمات البرمجية التي تم إنشاؤها بواسطة الوكيل.
-
-> [!WARNING]
-> يمكن لـ LLM توليد شفرة برمجية عشوائية سيتم تنفيذها بعد ذلك: لا تقمب استدعاء أى دوال غير آمنة!
-
-### موجه النظام
-
-ينشئ الوكيل، أو بالأحرى LLM الذي يقود الوكيل، يولد مخرجات بناءً على موجه النظام. يمكن تخصيص موجه النظام وتصميمه للمهام المقصودة. على سبيل المثال، تحقق من موجه النظام لـ [`ReactCodeAgent`] (الإصدار أدناه مبسط قليلاً).
-
-```text
-You will be given a task to solve as best you can.
-You have access to the following tools:
-<<tool_descriptions>>
-
-To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
-
-At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
-During each intermediate step, you can use 'print()' to save whatever important information you will then need.
-These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
-
-In the end you have to return a final answer using the `final_answer` tool.
-
-Here are a few examples using notional tools:
----
-{examples}
-
-Above example were using notional tools that might not exist for you. You only have access to those tools:
-<<tool_names>>
-You also can perform computations in the python code you generate.
-
-Always provide a 'Thought:' and a 'Code:\n```py' sequence ending with '```<end_code>' sequence. You MUST provide at least the 'Code:' sequence to move forward.
-
-Remember to not perform too many operations in a single code block! You should split the task into intermediate code blocks.
-Print results at the end of each step to save the intermediate results. Then use final_answer() to return the final result.
-
-Remember to make sure that variables you use are all defined.
-
-Now Begin!
-```
-
-يتضمن موجه النظام:
-- *مقدمة* تشرح كيف يجب أن يتصرف الوكيل والأدوات التي يجب عليه استخدامها.
-- وصف لجميع الأدوات التي يتم تحديدها بواسطة رمز `<<tool_descriptions>>` الذي يتم استبداله ديناميكيًا في وقت التشغيل بالأدوات التي يحددها المستخدم أو يختارها.
-    - يأتي وصف الأداة من سمات الأداة، `name`، و`description`، و`inputs` و`output_type`، وقالب `jinja2` بسيط يمكنك تحسينه.
-- شكل المخرج المتوقع.
-
-يمكنك تحسين موجه النظام، على سبيل المثال، عن طريق إضافة شرح لتنسيق المخرجات.
-
-للحصول على أقصى قدر من المرونة، يمكنك الكتابة فوق قالب موجه النظام بالكامل عن طريق تمرير موجه مخصص كمعامل إلى معلمة `system_prompt`.
-
-```python
-from transformers import ReactJsonAgent
-from transformers.agents import PythonInterpreterTool
-
-agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_custom_prompt}")
-```
-
-> [!WARNING]
-> يرجى التأكد من تحديد سلسلة `<<tool_descriptions>>` في مكان ما في `template` حتى يكون الوكيل على علم 
-بالأدوات المتاحة.
-
-
-### فحص تشغيل الوكيل
-
-فيما يلي بعض السمات المفيدة لفحص ما حدث بعد التشغيل:
-- تخزن  `agent.logs` سجلات مفصلة للوكيل. في كل خطوة من تشغيل الوكيل، يتم تخزين كل شيء في قاموس إلحاقه بـ `agent.logs`.
-- تشغيل `agent.write_inner_memory_from_logs()` يخلق ذاكرة داخلية لسجلات الوكيل للنظام LLM لعرضها، كقائمة من رسائل الدردشة. تنتقل هذه الطريقة عبر كل خطوة من سجل الوكيل ولا تخزن سوى ما يهمها كرسالة: على سبيل المثال، سيحفظ موجه النظام والمهمة في رسائل منفصلة، ثم لكل خطوة سيخزن مخرج LLM كرسالة، ومخرج استدعاء الأداة كرسالة أخرى. استخدم هذا إذا كنت تريد عرضًا عامًا لما حدث - ولكن لن يتم نسخ كل سجل بواسطة هذه الطريقة.
-
-## الأدوات
-
-الأداة هي عبارة عن وظيفة أساسية يستخدمها الوكيل لتنفيذ مهمة محددة.
-
-يمكنك على سبيل المثال التحقق من [`PythonInterpreterTool`]: لديه اسم ووصف ووصف للمدخلات ونوع للمخرج، وطريقة `__call__` التي تقوم بتنفيذ المهمة المطلوبة.
-
-عند تهيئة الوكيل، يتم استخدام سمات الأداة لتوليد وصف للأداة يتم تضمينه في موجه النظام الخاص بالوكيل. يتيح هذا للوكيل معرفة الأدوات التي يمكنه استخدامها ولماذا.
-
-### صندوق الأدوات الافتراضي
-
-يأتي Transformers مع صندوق أدوات افتراضي لتمكين الوكلاء، والذي يمكنك إضافته إلى وكيلك عند التهيئة باستخدام معامل `add_base_tools = True`:
-
-- **الإجابة على أسئلة المستند**: الإجابة على سؤال حول المستند (مثل ملف PDF) بتنسيق صورة ([Donut](./model_doc/donut))
-- **الإجابة على أسئلة الصور**: الإجابة على سؤال حول صورة ([VILT](./model_doc/vilt))
-- **التحدث إلى النص**: قم بتفريغ الكلام إلى نص ([Whisper](./model_doc/whisper))
-- **النص إلى كلام**: تحويل النص إلى كلام ([SpeechT5](./model_doc/speecht5))
-- **الترجمة**: ترجمة جملة معينة من لغة المصدر إلى لغة الهدف.
-- **مفسر كود Python**: تشغيل كود Python الذي تم إنشاؤه بواسطة LLM في بيئة آمنة. لن يتم إضافة هذه الأداة إلى [`ReactJsonAgent`] إلا إذا استخدمت `add_base_tools=True`، نظرًا لأن الأدوات المستندة إلى التعليمات البرمجية يمكنها بالفعل تنفيذ كود Python
-لا تترجم النصوص الخاصة ولا الأكواد البرمجية ولا الروابط ولا رموز HTML وCSS:
-
-يمكنك استخدام أداة يدويًا عن طريق استدعاء دالة [`load_tool`] وتحديد مهمة لتنفيذها.
-
-```python
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### إنشاء أداة جديدة
-
-يمكنك إنشاء أداتك الخاصة لتغطية حالات الاستخدام التي لا تغطيها الأدوات الافتراضية من Hugging Face.
-على سبيل المثال، دعنا نقوم بإنشاء أداة تعرض النموذج الأكثر تنزيلًا لمهمة معينة من Hub.
-
-سوف نبدأ بالكود التالي.
-
-```python
-from huggingface_hub import list_models
-
-task = "text-classification"
-
-model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-print(model.id)
-```
-
-يمكن تحويل هذه الشيفرة إلى فئة ترث من الفئة العليا [`Tool`].
-
-تحتاج الأداة المخصصة إلى:
-
-- اسم `name`، والتي تمثل اسم الأداة نفسها. عادةً ما يصف الاسم وظيفتها. بما أن الكود يعيد النموذج الأكثر تنزيلًا لمهمة ما، فلنسمها `model_download_counter`.
-- تستخدم خاصية `description` لملء موجه نظام الوكيل.
-- خاصية `inputs`، والتي هي عبارة عن قاموس بمفاتيح "type" و"description". يحتوي على معلومات تساعد المفسر Python على اتخاذ خيارات مستنيرة بشأن المدخلات.
-- خاصية `output_type`، والتي تحدد نوع المخرج.
-- طريقة `forward` والتي تحتوي على الكود الذي سيتم تنفيذه للحصول على النتيجة النهائية.
-
-```python
-from transformers import Tool
-from huggingface_hub import list_models
-
-class HFModelDownloadsTool(Tool):
-    name = "model_download_counter"
-    description = (
-        "This is a tool that returns the most downloaded model of a given task on the Hugging Face Hub. "
-        "It returns the name of the checkpoint."
-    )
-
-    inputs = {
-        "task": {
-            "type": "text",
-            "description": "the task category (such as text-classification, depth-estimation, etc)",
-        }
-    }
-    output_type = "text"
-
-    def forward(self, task: str):
-        model = next(iter(list_models(filter=task, sort="downloads", direction=-1)))
-        return model.id
-```
-
-الآن بعد أن أصبحت فئة `HfModelDownloadsTool` المخصصة جاهزة، يمكنك حفظها في ملف باسم `model_downloads.py` واستيرادها للاستخدام.
-
-```python
-from model_downloads import HFModelDownloadsTool
-
-tool = HFModelDownloadsTool()
-```
-
-يمكنك أيضًا مشاركة أداتك المخصصة في Hub عن طريق استدعاء [`~Tool.push_to_hub`] على الأداة. تأكد من أنك قمت بإنشاء مستودع لها على Hub وأنك تستخدم رمز وصول للقراءة.
-
-```python
-tool.push_to_hub("{your_username}/hf-model-downloads")
-```
-
-قم بتحميل الأداة باستخدام دالة [`~Tool.load_tool`] ومررها إلى معلمة `tools` في الوكيل الخاص بك.
-
-```python
-from transformers import load_tool, CodeAgent
-
-model_download_tool = load_tool("m-ric/hf-model-downloads")
-agent = CodeAgent(tools=[model_download_tool], llm_engine=llm_engine)
-agent.run(
-    "Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?"
-)
-```
-
-ستحصل على ما يلي:
-
-```text
-======== New task ========
-Can you give me the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub?
-==== Agent is executing the code below:
-most_downloaded_model = model_download_counter(task="text-to-video")
-print(f"The most downloaded model for the 'text-to-video' task is {most_downloaded_model}.")
-====
-```
-
-والناتج:
-
-`"النموذج الأكثر تنزيلًا لمهمة `text-to-video` هو ByteDance/AnimateDiff-Lightning."`
-
-### إدارة صندوق أدوات الوكيل الخاص بك
-
-إذا كنت قد قمت بتهيئة وكيل، فمن غير الملائم إعادة تهيئته من البداية لإضافة أداة جديدة ترغب في استخدامها. باستخدام مكتبة Transformers، يمكنك إدارة صندوق أدوات الوكيل بإضافة أو استبدال أداة موجودة.
-
-دعنا نضيف الأداة `model_download_tool` إلى وكيل تم تهيئته مسبقًا باستخدام صندوق الأدوات الافتراضي.
-
-```python
-from transformers import CodeAgent
-
-agent = CodeAgent(tools=[], llm_engine=llm_engine, add_base_tools=True)
-agent.toolbox.add_tool(model_download_tool)
-```
-
-الآن يمكننا الاستفادة من الأداة الجديدة وأداة تحويل النص إلى كلام السابقة:
-
-```python
-    agent.run(
-        "Can you read out loud the name of the model that has the most downloads in the 'text-to-video' task on the Hugging Face Hub and return the audio?"
-    )
-```
-
-| **Audio**                                                                                                                                            |
-|------------------------------------------------------------------------------------------------------------------------------------------------------|
-| <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/damo.wav" type="audio/wav"/> |
-
-> [!WARNING]
-> احترس عند إضافة أدوات إلى وكيل يعمل بالفعل لأنه يمكن أن يؤثر على اختيار الأداة لصالح أداتك أو اختيار أداة أخرى غير المحددة بالفعل.
-
-استخدم طريقة `agent.toolbox.update_tool()` لاستبدال أداة موجودة في صندوق أدوات الوكيل.
-هذا مفيد إذا كانت أداتك الجديدة بديلاً مباشرًا للأداة الموجودة لأن الوكيل يعرف بالفعل كيفية تنفيذ تلك المهمة المحددة.
-تأكد فقط من اتباع الأداة الجديدة لنفس واجهة برمجة التطبيقات (API) للأداة المستبدلة أو قم بتكييف قالب موجه النظام لضمان تحديث جميع الأمثلة التي تستخدم الأداة المستبدلة.
-
-### استخدام مجموعة من الأدوات
-
-يمكنك الاستفادة من مجموعات الأدوات باستخدام كائن ToolCollection، مع تحديد مجموعة الأدوات التي تريد استخدامها.
-ثم قم بتمريرها كقائمة لتهيئة الوكيل الخاص بك، وبدء استخدامها!
-
-```py
-from transformers import ToolCollection, ReactCodeAgent
-
-image_tool_collection = ToolCollection(collection_slug="huggingface-tools/diffusion-tools-6630bb19a942c2306a2cdb6f")
-agent = ReactCodeAgent(tools=[*image_tool_collection.tools], add_base_tools=True)
-
-agent.run("Please draw me a picture of rivers and lakes.")
-```
-
-لتسريع البداية، يتم تحميل الأدوات فقط إذا استدعاها الوكيل.
-
-ستحصل على هذه الصورة:
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" />
-
-### استخدام gradio-tools
-
-[gradio-tools](https://github.com/freddyaboulton/gradio-tools) هي مكتبة قوية تتيح استخدام Hugging
-Face Spaces كأدوات. تدعم العديد من المساحات الموجودة بالإضافة إلى مساحات مخصصة.
-
-تدعم مكتبة Transformers `gradio_tools` باستخدام طريقة [`Tool.from_gradio`] في الفئة. على سبيل المثال، دعنا نستخدم [`StableDiffusionPromptGeneratorTool`](https://github.com/freddyaboulton/gradio-tools/blob/main/gradio_tools/tools/prompt_generator.py) من مجموعة أدوات `gradio-tools` لتحسين المطالبات لإنشاء صور أفضل.
-
-استورد وقم بتهيئة الأداة، ثم مررها إلى طريقة `Tool.from_gradio`:
-
-```python
-from gradio_tools import StableDiffusionPromptGeneratorTool
-from transformers import Tool, load_tool, CodeAgent
-
-gradio_prompt_generator_tool = StableDiffusionPromptGeneratorTool()
-prompt_generator_tool = Tool.from_gradio(gradio_prompt_generator_tool)
-```
-
-الآن يمكنك استخدامه مثل أي أداة أخرى. على سبيل المثال، دعنا نحسن الموجه `a rabbit wearing a space suit`.
-
-```python
-image_generation_tool = load_tool('huggingface-tools/text-to-image')
-agent = CodeAgent(tools=[prompt_generator_tool, image_generation_tool], llm_engine=llm_engine)
-
-agent.run(
-    "Improve this prompt, then generate an image of it.", prompt='A rabbit wearing a space suit'
-)
-```
-
-يستفيد النموذج بشكل كافٍ من الأداة:
-
-```text
-======== New task ========
-Improve this prompt, then generate an image of it.
-You have been provided with these initial arguments: {'prompt': 'A rabbit wearing a space suit'}.
-==== Agent is executing the code below:
-improved_prompt = StableDiffusionPromptGenerator(query=prompt)
-while improved_prompt == "QUEUE_FULL":
-    improved_prompt = StableDiffusionPromptGenerator(query=prompt)
-print(f"The improved prompt is {improved_prompt}.")
-image = image_generator(prompt=improved_prompt)
-====
-```
-
-قبل إنشاء الصورة أخيرًا:
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rabbit_spacesuit_flux.webp" />
-
-> [!WARNING]
-> تتطلب gradio-tools إدخالات وإخراجات *نصية* حتى عند العمل مع طرائق مختلفة مثل كائنات الصور والصوت. الإدخالات والإخراجات الصورية والصوتية غير متوافقة حاليًا.
-
-### استخدام أدوات LangChain
-
-نحن نحب Langchain ونعتقد أنها تحتوي على مجموعة أدوات قوية للغاية.
-لاستيراد أداة من LangChain، استخدم الطريقة `from_langchain()`.
-
-فيما يلي كيفية استخدامها لإعادة إنشاء نتيجة البحث في المقدمة باستخدام أداة بحث الويب LangChain.
-
-```python
-from langchain.agents import load_tools
-from transformers import Tool, ReactCodeAgent
-
-search_tool = Tool.from_langchain(load_tools(["serpapi"])[0])
-
-agent = ReactCodeAgent(tools=[search_tool])
-
-agent.run("How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?")
-```
-
-## واجهة Gradio
-
-يمكنك الاستفادة من `gradio.Chatbot` لعرض أفكار الوكيل الخاص بك باستخدام `stream_to_gradio`، إليك مثال:
-
-```py
-import gradio as gr
-from transformers import (
-    load_tool,
-    ReactCodeAgent,
-    HfEngine,
-    stream_to_gradio,
-)
-
-# Import tool from Hub
-image_generation_tool = load_tool("m-ric/text-to-image")
-
-llm_engine = HfEngine("meta-llama/Meta-Llama-3-70B-Instruct")
-
-# Initialize the agent with the image generation tool
-agent = ReactCodeAgent(tools=[image_generation_tool], llm_engine=llm_engine)
-
-
-def interact_with_agent(task):
-    messages = []
-    messages.append(gr.ChatMessage(role="user", content=task))
-    yield messages
-    for msg in stream_to_gradio(agent, task):
-        messages.append(msg)
-        yield messages + [
-            gr.ChatMessage(role="assistant", content="⏳ Task not finished yet!")
-        ]
-    yield messages
-
-
-with gr.Blocks() as demo:
-    text_input = gr.Textbox(lines=1, label="Chat Message", value="Make me a picture of the Statue of Liberty.")
-    submit = gr.Button("Run illustrator agent!")
-    chatbot = gr.Chatbot(
-        label="Agent",
-        type="messages",
-        avatar_images=(
-            None,
-            "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png",
-        ),
-    )
-    submit.click(interact_with_agent, [text_input], [chatbot])
-
-if __name__ == "__main__":
-    demo.launch()
-```

docs/source/ar/bertology.md

@@ -15,4 +15,4 @@
 - الوصول إلى جميع أوزان الانتباه لكل رأس في BERT/GPT/GPT-2،
 - استرجاع قيم ومشتقات  مخرجات الرأس لحساب درجة أهمية الرأس وحذفه كما هو موضح في https://arxiv.org/abs/1905.10650.
 
-ولمساعدتك على فهم واستخدام هذه الميزات بسهولة، أضفنا مثالًا برمجيًا محددًا: [bertology.py](https://github.com/huggingface/transformers/tree/main/examples/research_projects/bertology/run_bertology.py) أثناء استخراج المعلومات  وتقليص من نموذج تم تدريبه مسبقًا على GLUE.
+ولمساعدتك على فهم واستخدام هذه الميزات بسهولة، أضفنا مثالًا برمجيًا محددًا: [bertology.py](https://github.com/huggingface/transformers-research-projects/tree/main/bertology/run_bertology.py) أثناء استخراج المعلومات  وتقليص من نموذج تم تدريبه مسبقًا على GLUE.

docs/source/ar/gguf.md

@@ -77,7 +77,7 @@ model = AutoModelForCausalLM.from_pretrained(model_id, gguf_file=filename)
 
 الآن لديك إمكانية الوصول إلى النسخة الكامل غير المكممة للنموذج في بيئة PyTorch، حيث يمكنك دمجه مع مجموعة كبيرة من الأدوات الأخرى.
 
-لإعادة التحويل إلى ملف `gguf`، نوصي باستخدام ملف [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert-hf-to-gguf.py) من llama.cpp.
+لإعادة التحويل إلى ملف `gguf`، نوصي باستخدام ملف [`convert-hf-to-gguf.py`](https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py) من llama.cpp.
 
 فيما يلي كيفية إكمال البرنامج النصي أعلاه لحفظ النموذج وإعادة تصديره مرة أخرى إلى `gguf`:
 

docs/source/ar/run_scripts.md

@@ -2,7 +2,7 @@
 
 بالإضافة إلى دفاتر الملاحظات [notebooks](./notebooks) الخاصة بـ 🤗 Transformers، هناك أيضًا نصوص برمجية توضيحية تُظهر كيفية تدريب نموذج لمهمة باستخدام [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch) أو [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow) أو [JAX/Flax](https://github.com/huggingface/transformers/tree/main/examples/flax).
 
-كما ستجد النصوص البرمجية التي استخدمناها في [مشاريع الأبحاث](https://github.com/huggingface/transformers/tree/main/examples/research_projects) و [الأمثلة القديمة](https://github.com/huggingface/transformers/tree/main/examples/legacy) والتي ساهم بها المجتمع بشكل أساسي. هذه النصوص البرمجية غير مدعومة بشكل نشط وقد تتطلب إصدارًا محددًا من مكتبة 🤗 Transformers والذي من المحتمل أن يكون غير متوافق مع الإصدار الأحدث من المكتبة.
+كما ستجد النصوص البرمجية التي استخدمناها في [مشاريع الأبحاث](https://github.com/huggingface/transformers-research-projects/) و [الأمثلة القديمة](https://github.com/huggingface/transformers/tree/main/examples/legacy) والتي ساهم بها المجتمع بشكل أساسي. هذه النصوص البرمجية غير مدعومة بشكل نشط وقد تتطلب إصدارًا محددًا من مكتبة 🤗 Transformers والذي من المحتمل أن يكون غير متوافق مع الإصدار الأحدث من المكتبة.
 
 لا يُتوقع أن تعمل النصوص البرمجية التوضيحية بشكل مباشر على كل مشكلة، وقد تحتاج إلى تكييف النص البرمجي مع المشكلة التي تحاول حلها. ولمساعدتك في ذلك، تعرض معظم النصوص البرمجية كيفية معالجة البيانات قبل التدريب بشكل كامل، مما يتيح لك تحريرها حسب الحاجة لحالتك الاستخدام.
 

docs/source/ar/trainer.md

@@ -674,29 +674,7 @@ use_cpu: false
 ```
 
 </hfoption>
-<hfoption id="Tensor Parallelism with PyTorch 2">
 
-```yml
-compute_environment: LOCAL_MACHINE
-tp_config:
-  tp_size: 4
-distributed_type: TP
-downcast_bf16: 'no'
-machine_rank: 0
-main_training_function: main
-mixed_precision: 'no'
-num_machines: 1
-num_processes: 4
-rdzv_backend: static
-same_network: true
-tpu_env: []
-tpu_use_cluster: false
-tpu_use_sudo: false
-use_cpu: false
-
-```
-
-</hfoption>
 </hfoptions>
 يُعد أمر  [`accelerate_launch`](https://huggingface.co/docs/accelerate/package_reference/cli#accelerate-launch) هو الطريقة المُوصى بها لتشغيل نص البرمجى للتدريب على نظام موزع باستخدام Accelerate و [`Trainer`] مع المعلمات المحددة في `config_file.yaml`. يتم حفظ هذا الملف في مجلد ذاكرة التخزين المؤقت لـ Accelerate ويتم تحميله تلقائيًا عند تشغيل `accelerate_launch`.
 

docs/source/de/_toctree.yml

@@ -23,8 +23,6 @@
     title: Laden und Trainieren von Adaptern mit 🤗 PEFT
   - local: model_sharing
     title: Ein Modell teilen
-  - local: transformers_agents
-    title: Agents
   - local: llm_tutorial
     title: Generation with LLMs
   title: Tutorials
@@ -39,4 +37,4 @@
     title: Testen
   - local: pr_checks
     title: Überprüfung einer Pull Request
-  title: Contribute
+  title: Contribute

docs/source/de/index.md

@@ -88,7 +88,7 @@ Die Bibliothek enthält derzeit JAX-, PyTorch- und TensorFlow-Implementierungen,
 1. **[DeiT](model_doc/deit)** (from Facebook) released with the paper [Training data-efficient image transformers & distillation through attention](https://arxiv.org/abs/2012.12877) by Hugo Touvron, Matthieu Cord, Matthijs Douze, Francisco Massa, Alexandre Sablayrolles, Hervé Jégou.
 1. **[DETR](model_doc/detr)** (from Facebook) released with the paper [End-to-End Object Detection with Transformers](https://arxiv.org/abs/2005.12872) by Nicolas Carion, Francisco Massa, Gabriel Synnaeve, Nicolas Usunier, Alexander Kirillov, Sergey Zagoruyko.
 1. **[DialoGPT](model_doc/dialogpt)** (from Microsoft Research) released with the paper [DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation](https://arxiv.org/abs/1911.00536) by Yizhe Zhang, Siqi Sun, Michel Galley, Yen-Chun Chen, Chris Brockett, Xiang Gao, Jianfeng Gao, Jingjing Liu, Bill Dolan.
-1. **[DistilBERT](model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation) and a German version of DistilBERT.
+1. **[DistilBERT](model_doc/distilbert)** (from HuggingFace), released together with the paper [DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108) by Victor Sanh, Lysandre Debut and Thomas Wolf. The same method has been applied to compress GPT2 into [DistilGPT2](https://github.com/huggingface/transformers-research-projects/tree/main/distillation), RoBERTa into [DistilRoBERTa](https://github.com/huggingface/transformers-research-projects/tree/main/distillation), Multilingual BERT into [DistilmBERT](https://github.com/huggingface/transformers-research-projects/tree/main/distillation) and a German version of DistilBERT.
 1. **[DiT](model_doc/dit)** (from Microsoft Research) released with the paper [DiT: Self-supervised Pre-training for Document Image Transformer](https://arxiv.org/abs/2203.02378) by Junlong Li, Yiheng Xu, Tengchao Lv, Lei Cui, Cha Zhang, Furu Wei.
 1. **[DPR](model_doc/dpr)** (from Facebook) released with the paper [Dense Passage Retrieval for Open-Domain Question Answering](https://arxiv.org/abs/2004.04906) by Vladimir Karpukhin, Barlas Oğuz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih.
 1. **[DPT](master/model_doc/dpt)** (from Intel Labs) released with the paper [Vision Transformers for Dense Prediction](https://arxiv.org/abs/2103.13413) by René Ranftl, Alexey Bochkovskiy, Vladlen Koltun.

docs/source/de/quicktour.md

@@ -156,7 +156,7 @@ Die [`pipeline`] kann jedes Modell aus dem [Model Hub](https://huggingface.co/mo
 
 <frameworkcontent>
 <pt>
-Use the [`AutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and it's associated tokenizer (more on an `AutoClass` below):
+Use the [`AutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and its associated tokenizer (more on an `AutoClass` below):
 
 ```py
 >>> from transformers import AutoTokenizer, AutoModelForSequenceClassification
@@ -166,7 +166,7 @@ Use the [`AutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the
 ```
 </pt>
 <tf>
-Use the [`TFAutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and it's associated tokenizer (more on an `TFAutoClass` below):
+Use the [`TFAutoModelForSequenceClassification`] and [`AutoTokenizer`] to load the pretrained model and its associated tokenizer (more on an `TFAutoClass` below):
 
 ```py
 >>> from transformers import AutoTokenizer, TFAutoModelForSequenceClassification
@@ -222,7 +222,7 @@ Anschließend wandelt der Tokenizer die Token in Zahlen um, um einen Tensor als
 Der Tokenizer gibt ein Wörterbuch zurück, das Folgendes enthält:
 
 * [input_ids](./glossary#input-ids): numerische Repräsentationen Ihrer Token.
-* [atttention_mask](.glossary#attention-mask): gibt an, welche Token beachtet werden sollen.
+* [attention_mask](.glossary#attention-mask): gibt an, welche Token beachtet werden sollen.
 
 Genau wie die [`pipeline`] akzeptiert der Tokenizer eine Liste von Eingaben. Darüber hinaus kann der Tokenizer den Text auch auffüllen und kürzen, um einen Stapel mit einheitlicher Länge zurückzugeben:
 

docs/source/de/run_scripts.md

@@ -18,7 +18,7 @@ rendered properly in your Markdown viewer.
 
 Neben den 🤗 Transformers [notebooks](./notebooks) gibt es auch Beispielskripte, die zeigen, wie man ein Modell für eine Aufgabe mit [PyTorch](https://github.com/huggingface/transformers/tree/main/examples/pytorch), [TensorFlow](https://github.com/huggingface/transformers/tree/main/examples/tensorflow) oder [JAX/Flax](https://github.com/huggingface/transformers/tree/main/examples/flax) trainiert.
 
-Sie werden auch Skripte finden, die wir in unseren [Forschungsprojekten](https://github.com/huggingface/transformers/tree/main/examples/research_projects) und [Legacy-Beispielen](https://github.com/huggingface/transformers/tree/main/examples/legacy) verwendet haben und die größtenteils von der Community stammen. Diese Skripte werden nicht aktiv gepflegt und erfordern eine bestimmte Version von 🤗 Transformers, die höchstwahrscheinlich nicht mit der neuesten Version der Bibliothek kompatibel ist.
+Sie werden auch Skripte finden, die wir in unseren [Forschungsprojekten](https://github.com/huggingface/transformers-research-projects/) und [Legacy-Beispielen](https://github.com/huggingface/transformers/tree/main/examples/legacy) verwendet haben und die größtenteils von der Community stammen. Diese Skripte werden nicht aktiv gepflegt und erfordern eine bestimmte Version von 🤗 Transformers, die höchstwahrscheinlich nicht mit der neuesten Version der Bibliothek kompatibel ist.
 
 Es wird nicht erwartet, dass die Beispielskripte bei jedem Problem sofort funktionieren. Möglicherweise müssen Sie das Skript an das Problem anpassen, das Sie zu lösen versuchen. Um Ihnen dabei zu helfen, legen die meisten Skripte vollständig offen, wie die Daten vorverarbeitet werden, so dass Sie sie nach Bedarf für Ihren Anwendungsfall bearbeiten können.
 

docs/source/de/transformers_agents.md

@@ -1,323 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Transformers Agents
-
-<Tip warning={true}>
-
-Transformers Agents ist eine experimentelle API, die jederzeit geändert werden kann. Die von den Agenten zurückgegebenen Ergebnisse
-zurückgegeben werden, können variieren, da sich die APIs oder die zugrunde liegenden Modelle ändern können.
-
-</Tip>
-
-Transformers Version v4.29.0, die auf dem Konzept von *Tools* und *Agenten* aufbaut. Sie können damit spielen in
-[dieses Colab](https://colab.research.google.com/drive/1c7MHD-T1forUPGcC_jlwsIptOzpG3hSj).
-
-Kurz gesagt, es bietet eine API für natürliche Sprache auf der Grundlage von Transformers: Wir definieren eine Reihe von kuratierten Tools und entwerfen einen 
-Agenten, um natürliche Sprache zu interpretieren und diese Werkzeuge zu verwenden. Es ist von vornherein erweiterbar; wir haben einige relevante Tools kuratiert, 
-aber wir werden Ihnen zeigen, wie das System einfach erweitert werden kann, um jedes von der Community entwickelte Tool zu verwenden.
-
-Beginnen wir mit einigen Beispielen dafür, was mit dieser neuen API erreicht werden kann. Sie ist besonders leistungsfähig, wenn es um 
-Sie ist besonders leistungsstark, wenn es um multimodale Aufgaben geht. Lassen Sie uns also eine Runde drehen, um Bilder zu erzeugen und Text vorzulesen.
-
-```py
-agent.run("Caption the following image", image=image)
-```
-
-| **Input**                                                                                                                   | **Output**                        |
-|-----------------------------------------------------------------------------------------------------------------------------|-----------------------------------|
-| <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/beaver.png" width=200> | A beaver is swimming in the water |
-
----
-
-```py
-agent.run("Read the following text out loud", text=text)
-```
-| **Input**                                                                                                               | **Output**                                   |
-|-------------------------------------------------------------------------------------------------------------------------|----------------------------------------------|
-| A beaver is swimming in the water | <audio controls><source src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tts_example.wav" type="audio/wav"> your browser does not support the audio element. </audio>
-
----
-
-```py
-agent.run(
-    "In the following `document`, where will the TRRF Scientific Advisory Council Meeting take place?",
-    document=document,
-)
-```
-| **Input**                                                                                                                   | **Output**     |
-|-----------------------------------------------------------------------------------------------------------------------------|----------------|
-| <img src="https://datasets-server.huggingface.co/assets/hf-internal-testing/example-documents/--/hf-internal-testing--example-documents/test/0/image/image.jpg" width=200> | ballroom foyer |
-
-## Schnellstart
-
-Bevor Sie `agent.run` verwenden können, müssen Sie einen Agenten instanziieren, der ein großes Sprachmodell (LLM) ist. 
-Wir bieten Unterstützung für openAI-Modelle sowie für OpenSource-Alternativen von BigCode und OpenAssistant. Die openAI
-Modelle sind leistungsfähiger (erfordern aber einen openAI-API-Schlüssel, können also nicht kostenlos verwendet werden); Hugging Face
-bietet kostenlosen Zugang zu Endpunkten für BigCode- und OpenAssistant-Modelle.
-
-To start with, please install the `agents` extras in order to install all default dependencies.
-```bash
-pip install transformers[agents]
-```
-
-Um openAI-Modelle zu verwenden, instanziieren Sie einen [`OpenAiAgent`], nachdem Sie die `openai`-Abhängigkeit installiert haben:
-
-```bash
-pip install openai
-```
-
-
-```py
-from transformers import OpenAiAgent
-
-agent = OpenAiAgent(model="text-davinci-003", api_key="<your_api_key>")
-```
-
-Um BigCode oder OpenAssistant zu verwenden, melden Sie sich zunächst an, um Zugriff auf die Inference API zu erhalten:
-
-```py
-from huggingface_hub import login
-
-login("<YOUR_TOKEN>")
-```
-
-Dann instanziieren Sie den Agenten
-
-```py
-from transformers import HfAgent
-
-# Starcoder
-agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoder")
-# StarcoderBase
-# agent = HfAgent("https://api-inference.huggingface.co/models/bigcode/starcoderbase")
-# OpenAssistant
-# agent = HfAgent(url_endpoint="https://api-inference.huggingface.co/models/OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5")
-```
-
-Dies geschieht mit der Inferenz-API, die Hugging Face derzeit kostenlos zur Verfügung stellt. Wenn Sie Ihren eigenen Inferenz
-Endpunkt für dieses Modell (oder einen anderen) haben, können Sie die obige URL durch Ihren URL-Endpunkt ersetzen.
-
-<Tip>
-
-StarCoder und OpenAssistant sind kostenlos und leisten bei einfachen Aufgaben bewundernswert gute Arbeit. Allerdings halten die Kontrollpunkte
-nicht, wenn es um komplexere Aufforderungen geht. Wenn Sie mit einem solchen Problem konfrontiert sind, empfehlen wir Ihnen, das OpenAI
-Modell auszuprobieren, das zwar leider nicht quelloffen ist, aber zur Zeit eine bessere Leistung erbringt.
-
-</Tip>
-
-Sie sind jetzt startklar! Lassen Sie uns in die beiden APIs eintauchen, die Ihnen jetzt zur Verfügung stehen.
-
-### Einzelne Ausführung (run)
-
-Die Methode der einmaligen Ausführung ist die Verwendung der [`~Agent.run`] Methode des Agenten:
-
-```py
-agent.run("Draw me a picture of rivers and lakes.")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200>
-
-Es wählt automatisch das (oder die) Werkzeug(e) aus, das (die) für die von Ihnen gewünschte Aufgabe geeignet ist (sind) und führt es (sie) entsprechend aus. Es
-kann eine oder mehrere Aufgaben in der gleichen Anweisung ausführen (je komplexer Ihre Anweisung ist, desto wahrscheinlicher ist ein
-der Agent scheitern).
-
-```py
-agent.run("Draw me a picture of the sea then transform the picture to add an island")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/sea_and_island.png" width=200>
-
-<br/>
-
-
-Jede [`~Agent.run`] Operation ist unabhängig, so dass Sie sie mehrmals hintereinander mit unterschiedlichen Aufgaben ausführen können.
-
-Beachten Sie, dass Ihr `Agent` nur ein großsprachiges Modell ist, so dass kleine Variationen in Ihrer Eingabeaufforderung völlig unterschiedliche Ergebnisse liefern können.
-unterschiedliche Ergebnisse liefern. Es ist wichtig, dass Sie die Aufgabe, die Sie ausführen möchten, so genau wie möglich erklären. Wir gehen noch weiter ins Detail
-wie man gute Prompts schreibt [hier](custom_tools#writing-good-user-inputs).
-
-Wenn Sie einen Status über Ausführungszeiten hinweg beibehalten oder dem Agenten Nicht-Text-Objekte übergeben möchten, können Sie dies tun, indem Sie
-Variablen, die der Agent verwenden soll. Sie könnten zum Beispiel das erste Bild von Flüssen und Seen erzeugen, 
-und das Modell bitten, dieses Bild zu aktualisieren und eine Insel hinzuzufügen, indem Sie Folgendes tun:
-
-```python
-picture = agent.run("Generate a picture of rivers and lakes.")
-updated_picture = agent.run("Transform the image in `picture` to add an island to it.", picture=picture)
-```
-
-<Tip>
-
-Dies kann hilfreich sein, wenn das Modell Ihre Anfrage nicht verstehen kann und die Werkzeuge verwechselt. Ein Beispiel wäre:
-
-```py
-agent.run("Draw me the picture of a capybara swimming in the sea")
-```
-
-Hier könnte das Modell auf zwei Arten interpretieren:
-- Die Funktion `Text-zu-Bild` erzeugt ein Wasserschwein, das im Meer schwimmt.
-- Oder Sie lassen das `Text-zu-Bild` ein Wasserschwein erzeugen und verwenden dann das Werkzeug `Bildtransformation`, um es im Meer schwimmen zu lassen.
-
-Falls Sie das erste Szenario erzwingen möchten, können Sie dies tun, indem Sie die Eingabeaufforderung als Argument übergeben:
-
-```py
-agent.run("Draw me a picture of the `prompt`", prompt="a capybara swimming in the sea")
-```
-
-</Tip>
-
-
-### Chat-basierte Ausführung (Chat)
-
-Der Agent verfügt auch über einen Chat-basierten Ansatz, der die Methode [`~Agent.chat`] verwendet:
-
-```py
-agent.chat("Generate a picture of rivers and lakes")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes.png" width=200> 
-
-```py
-agent.chat("Transform the picture so that there is a rock in there")
-```
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/rivers_and_lakes_and_beaver.png" width=200>
-
-<br/>
-
-Dies ist ein interessanter Ansatz, wenn Sie den Zustand über Anweisungen hinweg beibehalten möchten. Er ist besser für Experimente geeignet, 
-eignet sich aber eher für einzelne Anweisungen als für komplexe Anweisungen (die die [`~Agent.run`]
-Methode besser verarbeiten kann).
-
-Diese Methode kann auch Argumente entgegennehmen, wenn Sie Nicht-Text-Typen oder bestimmte Aufforderungen übergeben möchten.
-
-### ⚠️ Fernausführung
-
-Zu Demonstrationszwecken und damit es mit allen Setups verwendet werden kann, haben wir Remote-Executors für mehrere 
-der Standard-Tools erstellt, auf die der Agent in dieser Version Zugriff hat. Diese werden erstellt mit 
-[inference endpoints](https://huggingface.co/inference-endpoints).
-
-Wir haben diese vorerst deaktiviert, aber um zu sehen, wie Sie selbst Remote Executors Tools einrichten können,
-empfehlen wir die Lektüre des [custom tool guide](./custom_tools).
-
-### Was passiert hier? Was sind Tools und was sind Agenten?
-
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/diagram.png">
-
-#### Agenten
-
-Der "Agent" ist hier ein großes Sprachmodell, das wir auffordern, Zugang zu einem bestimmten Satz von Tools zu erhalten.
-
-LLMs sind ziemlich gut darin, kleine Codeproben zu erzeugen. Diese API macht sich das zunutze, indem sie das 
-LLM ein kleines Codebeispiel gibt, das eine Aufgabe mit einer Reihe von Werkzeugen ausführt. Diese Aufforderung wird dann ergänzt durch die 
-Aufgabe, die Sie Ihrem Agenten geben, und die Beschreibung der Werkzeuge, die Sie ihm geben. Auf diese Weise erhält er Zugriff auf die Dokumentation der 
-Tools, insbesondere die erwarteten Eingaben und Ausgaben, und kann den entsprechenden Code generieren.
-
-#### Tools
-
-Tools sind sehr einfach: Sie bestehen aus einer einzigen Funktion mit einem Namen und einer Beschreibung. Wir verwenden dann die Beschreibungen dieser Tools 
-um den Agenten aufzufordern. Anhand der Eingabeaufforderung zeigen wir dem Agenten, wie er die Tools nutzen kann, um das zu tun, was in der 
-in der Abfrage angefordert wurde.
-
-Dies geschieht mit brandneuen Tools und nicht mit Pipelines, denn der Agent schreibt besseren Code mit sehr atomaren Tools. 
-Pipelines sind stärker refaktorisiert und fassen oft mehrere Aufgaben in einer einzigen zusammen. Tools sind dafür gedacht, sich auf
-eine einzige, sehr einfache Aufgabe konzentrieren.
-
-#### Code-Ausführung?!
-
-Dieser Code wird dann mit unserem kleinen Python-Interpreter auf den mit Ihren Tools übergebenen Eingaben ausgeführt. 
-Wir hören Sie schon schreien "Willkürliche Codeausführung!", aber lassen Sie uns erklären, warum das nicht der Fall ist.
-
-Die einzigen Funktionen, die aufgerufen werden können, sind die von Ihnen zur Verfügung gestellten Tools und die Druckfunktion, so dass Sie bereits eingeschränkt sind 
-eingeschränkt, was ausgeführt werden kann. Sie sollten sicher sein, wenn es sich auf die Werkzeuge für das Umarmungsgesicht beschränkt. 
-
-Dann lassen wir keine Attributsuche oder Importe zu (die ohnehin nicht benötigt werden, um die 
-Inputs/Outputs an eine kleine Gruppe von Funktionen), so dass alle offensichtlichen Angriffe (und Sie müssten den LLM 
-dazu auffordern, sie auszugeben) kein Problem darstellen sollten. Wenn Sie auf Nummer sicher gehen wollen, können Sie die 
-run()-Methode mit dem zusätzlichen Argument return_code=True ausführen. In diesem Fall gibt der Agent nur den auszuführenden Code 
-zur Ausführung zurück und Sie können entscheiden, ob Sie ihn ausführen möchten oder nicht.
-
-Die Ausführung bricht bei jeder Zeile ab, in der versucht wird, eine illegale Operation auszuführen, oder wenn ein regulärer Python-Fehler 
-mit dem vom Agenten generierten Code.
-
-### Ein kuratierter Satz von Tools
-
-Wir haben eine Reihe von Tools identifiziert, die solche Agenten unterstützen können. Hier ist eine aktualisierte Liste der Tools, die wir integriert haben 
-in `transformers` integriert haben:
-
-- **Beantwortung von Fragen zu Dokumenten**: Beantworten Sie anhand eines Dokuments (z.B. PDF) im Bildformat eine Frage zu diesem Dokument ([Donut](./model_doc/donut))
-- Beantworten von Textfragen**: Geben Sie einen langen Text und eine Frage an, beantworten Sie die Frage im Text ([Flan-T5](./model_doc/flan-t5))
-- **Unbedingte Bildunterschriften**: Beschriften Sie das Bild! ([BLIP](./model_doc/blip))
-- **Bildfragebeantwortung**: Beantworten Sie bei einem Bild eine Frage zu diesem Bild ([VILT](./model_doc/vilt))
-- **Bildsegmentierung**: Geben Sie ein Bild und einen Prompt an und geben Sie die Segmentierungsmaske dieses Prompts aus ([CLIPSeg](./model_doc/clipseg))
-- **Sprache in Text**: Geben Sie eine Audioaufnahme einer sprechenden Person an und transkribieren Sie die Sprache in Text ([Whisper](./model_doc/whisper))
-- **Text in Sprache**: wandelt Text in Sprache um ([SpeechT5](./model_doc/speecht5))
-- **Zero-Shot-Textklassifizierung**: Ermitteln Sie anhand eines Textes und einer Liste von Bezeichnungen, welcher Bezeichnung der Text am ehesten entspricht ([BART](./model_doc/bart))
-- **Textzusammenfassung**: fassen Sie einen langen Text in einem oder wenigen Sätzen zusammen ([BART](./model_doc/bart))
-- **Übersetzung**: Übersetzen des Textes in eine bestimmte Sprache ([NLLB](./model_doc/nllb))
-
-Diese Tools sind in Transformatoren integriert und können auch manuell verwendet werden, zum Beispiel:
-
-```py
-from transformers import load_tool
-
-tool = load_tool("text-to-speech")
-audio = tool("This is a text to speech tool")
-```
-
-### Benutzerdefinierte Tools
-
-Wir haben zwar eine Reihe von Tools identifiziert, sind aber der festen Überzeugung, dass der Hauptwert dieser Implementierung darin besteht 
-die Möglichkeit, benutzerdefinierte Tools schnell zu erstellen und weiterzugeben.
-
-Indem Sie den Code eines Tools in einen Hugging Face Space oder ein Modell-Repository stellen, können Sie das Tool 
-direkt mit dem Agenten nutzen. Wir haben ein paar neue Funktionen hinzugefügt 
-**transformers-agnostic** Tools zur [`huggingface-tools` Organisation](https://huggingface.co/huggingface-tools) hinzugefügt:
-
-- **Text-Downloader**: zum Herunterladen eines Textes von einer Web-URL
-- **Text zu Bild**: erzeugt ein Bild nach einer Eingabeaufforderung und nutzt dabei stabile Diffusion
-- **Bildtransformation**: verändert ein Bild anhand eines Ausgangsbildes und einer Eingabeaufforderung, unter Ausnutzung der stabilen pix2pix-Diffusion
-- **Text zu Video**: Erzeugen eines kleinen Videos nach einer Eingabeaufforderung, unter Verwendung von damo-vilab
-
-Das Text-zu-Bild-Tool, das wir von Anfang an verwendet haben, ist ein Remote-Tool, das sich in 
-[*huggingface-tools/text-to-image*](https://huggingface.co/spaces/huggingface-tools/text-to-image)! Wir werden
-weiterhin solche Tools für diese und andere Organisationen veröffentlichen, um diese Implementierung weiter zu verbessern.
-
-Die Agenten haben standardmäßig Zugriff auf die Tools, die sich auf [*huggingface-tools*](https://huggingface.co/huggingface-tools) befinden.
-Wie Sie Ihre eigenen Tools schreiben und freigeben können und wie Sie jedes benutzerdefinierte Tool, das sich auf dem Hub befindet, nutzen können, erklären wir in [folgender Anleitung](custom_tools).
-
-### Code-Erzeugung
-
-Bisher haben wir gezeigt, wie Sie die Agenten nutzen können, um Aktionen für Sie durchzuführen. Der Agent generiert jedoch nur Code
-den wir dann mit einem sehr eingeschränkten Python-Interpreter ausführen. Falls Sie den generierten Code in einer anderen Umgebung verwenden möchten 
-einer anderen Umgebung verwenden möchten, können Sie den Agenten auffordern, den Code zusammen mit einer Tooldefinition und genauen Importen zurückzugeben.
-
-Zum Beispiel die folgende Anweisung
-```python
-agent.run("Draw me a picture of rivers and lakes", return_code=True)
-```
-
-gibt den folgenden Code zurück
-
-```python
-from transformers import load_tool
-
-image_generator = load_tool("huggingface-tools/text-to-image")
-
-image = image_generator(prompt="rivers and lakes")
-```
-
-die Sie dann selbst ändern und ausführen können.

docs/source/en/_toctree.yml

@@ -1,16 +1,14 @@
-- title: Get started
-  sections:
+- sections:
   - local: index
     title: Transformers
   - local: installation
     title: Installation
   - local: quicktour
     title: Quickstart
-- title: Base classes
-  isExpanded: False
+  title: Get started
+- isExpanded: false
   sections:
-  - title: Models
-    sections:
+  - sections:
     - local: models
       title: Loading models
     - local: custom_models
@@ -31,8 +29,10 @@
       title: The Transformer model family
     - local: attention
       title: Attention mechanisms
-  - title: Preprocessors
-    sections:
+    - local: attention_interface
+      title: Customizing attention function
+    title: Models
+  - sections:
     - local: fast_tokenizers
       title: Tokenizers
     - local: image_processors
@@ -47,11 +47,11 @@
       title: Summary of the tokenizers
     - local: pad_truncation
       title: Padding and truncation
-- title: Inference
-  isExpanded: False
+    title: Preprocessors
+  title: Base classes
+- isExpanded: false
   sections:
-  - title: Pipeline API
-    sections:
+  - sections:
     - local: pipeline_tutorial
       title: Pipeline
     - local: pipeline_gradio
@@ -60,8 +60,8 @@
       title: Web server inference
     - local: add_new_pipeline
       title: Adding a new pipeline
-  - title: LLMs
-    sections:
+    title: Pipeline API
+  - sections:
     - local: llm_tutorial
       title: Text generation
     - local: generation_strategies
@@ -74,14 +74,16 @@
       title: Optimizing inference
     - local: kv_cache
       title: KV cache strategies
+    - local: serving
+      title: Serving
     - local: cache_explanation
       title: Caching
     - local: llm_tutorial_optimization
       title: Getting the most out of LLMs
     - local: perplexity
       title: Perplexity of fixed-length models
-  - title: Chat with models
-    sections:
+    title: LLMs
+  - sections:
     - local: conversations
       title: Chat basics
     - local: chat_templating
@@ -92,8 +94,8 @@
       title: Template writing
     - local: chat_extras
       title: Tools and RAG
-  - title: Optimization
-    sections:
+    title: Chat with models
+  - sections:
     - local: perf_torch_compile
       title: torch.compile
     - local: perf_infer_gpu_one
@@ -104,15 +106,15 @@
       title: CPU
     - local: tf_xla
       title: XLA
+    title: Optimization
   - local: agents
     title: Agents
   - local: tools
     title: Tools
-- title: Training
-  isExpanded: False
+  title: Inference
+- isExpanded: false
   sections:
-  - title: Trainer API
-    sections:
+  - sections:
     - local: trainer
       title: Trainer
     - local: training
@@ -121,8 +123,8 @@
       title: Optimizers
     - local: hpo_train
       title: Hyperparameter search
-  - title: Distributed training
-    sections:
+    title: Trainer API
+  - sections:
     - local: gpu_selection
       title: GPU selection
     - local: accelerate
@@ -137,8 +139,8 @@
       title: Distributed CPUs
     - local: perf_train_gpu_many
       title: Parallelism methods
-  - title: Hardware
-    sections:
+    title: Distributed training
+  - sections:
     - local: perf_train_gpu_one
       title: GPU
     - local: perf_train_cpu
@@ -149,17 +151,24 @@
       title: Apple Silicon
     - local: perf_hardware
       title: Build your own machine
+    title: Hardware
   - local: peft
     title: PEFT
   - local: model_memory_anatomy
     title: Model training anatomy
-- title: Quantization
-  isExpanded: False
+  title: Training
+- isExpanded: false
   sections:
   - local: quantization/overview
     title: Overview
+  - local: quantization/selecting
+    title: Selecting a quantization method
+  - local: quantization/concept_guide
+    title: Quantization concepts
   - local: quantization/aqlm
     title: AQLM
+  - local: quantization/auto_round
+    title: AutoRound
   - local: quantization/awq
     title: AWQ
   - local: quantization/bitnet
@@ -186,6 +195,8 @@
     title: Optimum
   - local: quantization/quanto
     title: Quanto
+  - local: quantization/quark
+    title: Quark
   - local: quantization/torchao
     title: torchao
   - local: quantization/spqr
@@ -194,8 +205,8 @@
     title: VPTQ
   - local: quantization/contribute
     title: Contribute
-- title: Export to production
-  isExpanded: False
+  title: Quantization
+- isExpanded: false
   sections:
   - local: serialization
     title: ONNX
@@ -205,13 +216,11 @@
     title: ExecuTorch
   - local: torchscript
     title: TorchScript
-- title: Resources
-  isExpanded: False
+  title: Export to production
+- isExpanded: false
   sections:
-  - title: Task recipes
-    sections:
-    - title: Natural language processing
-      sections:
+  - sections:
+    - sections:
       - local: tasks/sequence_classification
         title: Text classification
       - local: tasks/token_classification
@@ -228,14 +237,14 @@
         title: Summarization
       - local: tasks/multiple_choice
         title: Multiple choice
-    - title: Audio
-      sections:
+      title: Natural language processing
+    - sections:
       - local: tasks/audio_classification
         title: Audio classification
       - local: tasks/asr
         title: Automatic speech recognition
-    - title: Computer vision
-      sections:
+      title: Audio
+    - sections:
       - local: tasks/image_classification
         title: Image classification
       - local: tasks/semantic_segmentation
@@ -260,8 +269,8 @@
         title: Keypoint detection
       - local: tasks/knowledge_distillation_for_image_classification
         title: Knowledge Distillation for Computer Vision
-    - title: Multimodal
-      sections:
+      title: Computer vision
+    - sections:
       - local: tasks/image_captioning
         title: Image captioning
       - local: tasks/document_question_answering
@@ -276,6 +285,10 @@
         title: Image-text-to-text
       - local: tasks/video_text_to_text
         title: Video-text-to-text
+      - local: tasks/visual_document_retrieval
+        title: Visual Document Retrieval
+      title: Multimodal
+    title: Task recipes
   - local: run_scripts
     title: Training scripts
   - local: glossary
@@ -288,8 +301,8 @@
     title: Community resources
   - local: troubleshooting
     title: Troubleshoot
-- title: Contribute
-  isExpanded: False
+  title: Resources
+- isExpanded: false
   sections:
   - local: contributing
     title: Contribute to Transformers
@@ -297,13 +310,10 @@
     title: Transformers model tests
   - local: pr_checks
     title: Pull request checks
-- title: API
-  isExpanded: False
+  title: Contribute
+- isExpanded: false
   sections:
-  - title: Main classes
-    sections:
-    - local: main_classes/agent
-      title: Agents and Tools
+  - sections:
     - local: model_doc/auto
       title: Auto Classes
     - local: main_classes/backbones
@@ -348,10 +358,9 @@
       title: Feature Extractor
     - local: main_classes/image_processor
       title: Image Processor
-  - title: Models
-    sections:
-    - title: Text models
-      sections:
+    title: Main classes
+  - sections:
+    - sections:
       - local: model_doc/albert
         title: ALBERT
       - local: model_doc/bamba
@@ -376,6 +385,8 @@
         title: BigBirdPegasus
       - local: model_doc/biogpt
         title: BioGpt
+      - local: model_doc/bitnet
+        title: BitNet
       - local: model_doc/blenderbot
         title: Blenderbot
       - local: model_doc/blenderbot-small
@@ -412,6 +423,8 @@
         title: DeBERTa
       - local: model_doc/deberta-v2
         title: DeBERTa-v2
+      - local: model_doc/deepseek_v3
+        title: DeepSeek-V3
       - local: model_doc/dialogpt
         title: DialoGPT
       - local: model_doc/diffllama
@@ -456,6 +469,8 @@
         title: Gemma2
       - local: model_doc/glm
         title: GLM
+      - local: model_doc/glm4
+        title: glm4
       - local: model_doc/openai-gpt
         title: GPT
       - local: model_doc/gpt_neo
@@ -480,8 +495,6 @@
         title: GraniteMoe
       - local: model_doc/granitemoeshared
         title: GraniteMoeShared
-      - local: model_doc/granitevision
-        title: GraniteVision
       - local: model_doc/helium
         title: Helium
       - local: model_doc/herbert
@@ -594,6 +607,10 @@
         title: Qwen2
       - local: model_doc/qwen2_moe
         title: Qwen2MoE
+      - local: model_doc/qwen3
+        title: Qwen3
+      - local: model_doc/qwen3_moe
+        title: Qwen3MoE
       - local: model_doc/rag
         title: RAG
       - local: model_doc/realm
@@ -660,8 +677,8 @@
         title: Zamba
       - local: model_doc/zamba2
         title: Zamba2
-    - title: Vision models
-      sections:
+      title: Text models
+    - sections:
       - local: model_doc/beit
         title: BEiT
       - local: model_doc/bit
@@ -720,6 +737,8 @@
         title: Mask2Former
       - local: model_doc/maskformer
         title: MaskFormer
+      - local: model_doc/mlcd
+        title: MLCD
       - local: model_doc/mobilenet_v1
         title: MobileNetV1
       - local: model_doc/mobilenet_v2
@@ -732,6 +751,8 @@
         title: NAT
       - local: model_doc/poolformer
         title: PoolFormer
+      - local: model_doc/prompt_depth_anything
+        title: Prompt Depth Anything
       - local: model_doc/pvt
         title: Pyramid Vision Transformer (PVT)
       - local: model_doc/pvt_v2
@@ -788,8 +809,8 @@
         title: YOLOS
       - local: model_doc/zoedepth
         title: ZoeDepth
-    - title: Audio models
-      sections:
+      title: Vision models
+    - sections:
       - local: model_doc/audio-spectrogram-transformer
         title: Audio Spectrogram Transformer
       - local: model_doc/bark
@@ -802,6 +823,8 @@
         title: EnCodec
       - local: model_doc/fastspeech2_conformer
         title: FastSpeech2Conformer
+      - local: model_doc/granite_speech
+        title: GraniteSpeech
       - local: model_doc/hubert
         title: Hubert
       - local: model_doc/mctct
@@ -858,16 +881,16 @@
         title: XLS-R
       - local: model_doc/xlsr_wav2vec2
         title: XLSR-Wav2Vec2
-    - title: Video models
-      sections:
+      title: Audio models
+    - sections:
       - local: model_doc/timesformer
         title: TimeSformer
       - local: model_doc/videomae
         title: VideoMAE
       - local: model_doc/vivit
         title: ViViT
-    - title: Multimodal models
-      sections:
+      title: Video models
+    - sections:
       - local: model_doc/align
         title: ALIGN
       - local: model_doc/altclip
@@ -906,10 +929,14 @@
         title: Emu3
       - local: model_doc/flava
         title: FLAVA
+      - local: model_doc/gemma3
+        title: Gemma3
       - local: model_doc/git
         title: GIT
       - local: model_doc/got_ocr2
         title: GOT-OCR2
+      - local: model_doc/granitevision
+        title: GraniteVision
       - local: model_doc/grounding-dino
         title: Grounding DINO
       - local: model_doc/groupvit
@@ -924,6 +951,10 @@
         title: InstructBLIP
       - local: model_doc/instructblipvideo
         title: InstructBlipVideo
+      - local: model_doc/internvl
+        title: InternVL
+      - local: model_doc/janus
+        title: Janus
       - local: model_doc/kosmos-2
         title: KOSMOS-2
       - local: model_doc/layoutlm
@@ -936,6 +967,8 @@
         title: LayoutXLM
       - local: model_doc/lilt
         title: LiLT
+      - local: model_doc/llama4
+        title: Llama4
       - local: model_doc/llava
         title: Llava
       - local: model_doc/llava_next
@@ -950,6 +983,8 @@
         title: MatCha
       - local: model_doc/mgp-str
         title: MGP-STR
+      - local: model_doc/mistral3
+        title: Mistral3
       - local: model_doc/mllama
         title: mllama
       - local: model_doc/nougat
@@ -966,10 +1001,14 @@
         title: PaliGemma
       - local: model_doc/perceiver
         title: Perceiver
+      - local: model_doc/phi4_multimodal
+        title: Phi4 Multimodal
       - local: model_doc/pix2struct
         title: Pix2Struct
       - local: model_doc/pixtral
         title: Pixtral
+      - local: model_doc/qwen2_5_omni
+        title: Qwen2.5-Omni
       - local: model_doc/qwen2_5_vl
         title: Qwen2.5-VL
       - local: model_doc/qwen2_audio
@@ -978,6 +1017,8 @@
         title: Qwen2VL
       - local: model_doc/sam
         title: Segment Anything
+      - local: model_doc/shieldgemma2
+        title: ShieldGemma2
       - local: model_doc/siglip
         title: SigLIP
       - local: model_doc/siglip2
@@ -1010,14 +1051,14 @@
         title: VisualBERT
       - local: model_doc/xclip
         title: X-CLIP
-    - title: Reinforcement learning models
-      sections:
+      title: Multimodal models
+    - sections:
       - local: model_doc/decision_transformer
         title: Decision Transformer
       - local: model_doc/trajectory_transformer
         title: Trajectory Transformer
-    - title: Time series models
-      sections:
+      title: Reinforcement learning models
+    - sections:
       - local: model_doc/autoformer
         title: Autoformer
       - local: model_doc/informer
@@ -1028,14 +1069,19 @@
         title: PatchTST
       - local: model_doc/time_series_transformer
         title: Time Series Transformer
-    - title: Graph models
-      sections:
+      - local: model_doc/timesfm
+        title: TimesFM
+      title: Time series models
+    - sections:
       - local: model_doc/graphormer
         title: Graphormer
-  - title: Internal helpers
-    sections:
+      title: Graph models
+    title: Models
+  - sections:
     - local: internal/modeling_utils
       title: Custom Layers and Utilities
+    - local: internal/model_debugging_utils
+      title: Utilities for Model Debugging
     - local: internal/pipelines_utils
       title: Utilities for pipelines
     - local: internal/tokenization_utils
@@ -1050,6 +1096,9 @@
       title: Utilities for Audio processing
     - local: internal/file_utils
       title: General Utilities
+    - local: internal/import_utils
+      title: Importing Utilities
     - local: internal/time_series_utils
       title: Utilities for Time Series
-      
+    title: Internal helpers
+  title: API

docs/source/en/add_new_model.md

@@ -476,7 +476,7 @@ When both implementations produce the same output, verify the outputs are within
 torch.allclose(original_output, output, atol=1e-3)
 ```
 
-This is typically the most difficult part of the process. Congratulations if you've made it this far! 
+This is typically the most difficult part of the process. Congratulations if you've made it this far!
 
 And if you're stuck or struggling with this step, don't hesitate to ask for help on your pull request.
 
@@ -541,6 +541,48 @@ input_ids = tokenizer(input_str).input_ids
 
 When both implementations have the same `input_ids`, add a tokenizer test file. This file is analogous to the modeling test files. The tokenizer test files should contain a couple of hardcoded integration tests.
 
+## Implement image processor
+
+> [!TIP]
+> Fast image processors use the [torchvision](https://pytorch.org/vision/stable/index.html) library and can perform image processing on the GPU, significantly improving processing speed.
+> We recommend adding a fast image processor ([`BaseImageProcessorFast`]) in addition to the "slow" image processor ([`BaseImageProcessor`]) to provide users with the best performance. Feel free to tag [@yonigozlan](https://github.com/yonigozlan) for help adding a [`BaseImageProcessorFast`].
+
+While this example doesn't include an image processor, you may need to implement one if your model requires image inputs. The image processor is responsible for converting images into a format suitable for your model. Before implementing a new one, check whether an existing image processor in the Transformers library can be reused, as many models share similar image processing techniques. Note that you can also use [modular](./modular_transformers) for image processors to reuse existing components.
+
+If you do need to implement a new image processor, refer to an existing image processor to understand the expected structure. Slow image processors ([`BaseImageProcessor`]) and fast image processors ([`BaseImageProcessorFast`]) are designed differently, so make sure you follow the correct structure based on the processor type you're implementing.
+
+Run the following command (only if you haven't already created the fast image processor with the `transformers-cli add-new-model-like` command) to generate the necessary imports and to create a prefilled template for the fast image processor. Modify the template to fit your model.
+
+```bash
+transformers-cli add-fast-image-processor --model-name your_model_name
+```
+
+This command will generate the necessary imports and provide a pre-filled template for the fast image processor. You can then modify it to fit your model's needs.
+
+Add tests for the image processor in `tests/models/your_model_name/test_image_processing_your_model_name.py`. These tests should be similar to those for other image processors and should verify that the image processor correctly handles image inputs. If your image processor includes unique features or processing methods, ensure you add specific tests for those as well.
+
+## Implement processor
+
+If your model accepts multiple modalities, like text and images, you need to add a processor. The processor centralizes the preprocessing of different modalities before passing them to the model.
+
+The processor should call the appropriate modality-specific processors within its `__call__` function to handle each type of input correctly. Be sure to check existing processors in the library to understand their expected structure. Transformers uses the following convention in the `__call__` function signature.
+
+```python
+def __call__(
+    self,
+    images: ImageInput = None,
+    text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+    audio=None,
+    videos=None,
+    **kwargs: Unpack[YourModelProcessorKwargs],
+) -> BatchFeature:
+    ...
+```
+
+`YourModelProcessorKwargs` is a `TypedDict` that includes all the typical processing arguments and any extra arguments a specific processor may require.
+
+Add tests for the processor in `tests/models/your_model_name/test_processor_your_model_name.py`. These tests should be similar to those for other processors and should verify that the processor correctly handles the different modalities.
+
 ## Integration tests
 
 Now that you have a model and tokenizer, add end-to-end integration tests for the model and tokenizer to `tests/models/brand_new_llama/test_modeling_brand_new_llama.py`.
@@ -620,4 +662,4 @@ There are four timelines for model additions depending on the model contributor
 
 - **Hub-first release**: Transformers [remote-code](./models#custom-models) feature allows Transformers-based projects to be shared directly on the Hub. This is a good option if you don't have the bandwidth to add a model directly to Transformers.
 
-  If a model ends up being very popular, then it's very likely that we'll integrate it in Transformers ourselves to enable better support (documentation, maintenance, optimization, etc.) for it. A Hub-first release is the most frictionless way to add a model.
+  If a model ends up being very popular, then it's very likely that we'll integrate it in Transformers ourselves to enable better support (documentation, maintenance, optimization, etc.) for it. A Hub-first release is the most frictionless way to add a model.

docs/source/en/agents.md

@@ -15,283 +15,4 @@ rendered properly in your Markdown viewer.
 -->
 
 > [!WARNING]
-> Agents and tools are being spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. These docs will be deprecated in the future!
-
-# Agents
-
-[[open-in-colab]]
-
-An agent is a system where a large language model (LLM) can execute more complex tasks through *planning* and using *tools*.
-
-- Planning helps a LLM reason its way through a task by breaking it down into smaller subtasks. For example, [`CodeAgent`] plans a series of actions to take and then generates Python code to execute all the actions at once.
-
-    Another planning method is by self-reflection and refinement of its previous actions to improve its performance. The [`ReactJsonAgent`] is an example of this type of planning, and it's based on the [ReAct](https://hf.co/papers/2210.03629) framework. This agent plans and executes actions one at a time based on the feedback it receives from each action.
-
-- Tools give a LLM access to external functions or APIs that it can use to help it complete a task. For example, [gradio-tools](https://github.com/freddyaboulton/gradio-tools) gives a LLM access to any of the [Gradio](https://www.gradio.app/) apps available on Hugging Face [Spaces](https://hf.co/spaces). These apps can be used for a wide range of tasks such as image generation, video generation, audio transcription, and more.
-
-To use agents in Transformers, make sure you have the extra `agents` dependencies installed.
-
-```bash
-!pip install transformers[agents]
-```
-
-Create an agent instance (refer to the [Agents](./main_classes/agent#agents) API for supported agents in Transformers) and a list of tools available for it to use, then [`~ReactAgent.run`] the agent on your task. The example below demonstrates how a ReAct agent reasons through a task.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[])
-agent.run(
-    "How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?",
-)
-```
-
-```bash
-======== New task ========
-How many more blocks (also denoted as layers) in BERT base encoder than the encoder from the architecture proposed in Attention is All You Need?
-==== Agent is executing the code below:
-bert_layers = 12  # BERT base encoder has 12 layers
-attention_layers = 6  # Encoder in Attention is All You Need has 6 layers
-layer_diff = bert_layers - attention_layers
-print("The difference in layers between BERT base encoder and Attention is All You Need is", layer_diff)
-====
-Print outputs:
-The difference in layers between BERT base encoder and Attention is All You Need is 6
-
-==== Agent is executing the code below:
-final_answer("BERT base encoder has {} more layers than the encoder from Attention is All You Need.".format(layer_diff))
-====
-Print outputs:
-
->>> Final answer:
-BERT base encoder has 6 more layers than the encoder from Attention is All You Need.
-```
-
-This guide will walk you through in more detail how to initialize an agent.
-
-## LLM
-
-An agent uses a LLM to plan and execute a task; it is the engine that powers the agent. To choose and build your own LLM engine, you need a method that:
-
-1. the input uses the [chat template](./chat_templating) format, `List[Dict[str, str]]`, and it returns a string
-2. the LLM stops generating outputs when it encounters the sequences in `stop_sequences`
-
-```py
-def llm_engine(messages, stop_sequences=["Task"]) -> str:
-    response = client.chat_completion(messages, stop=stop_sequences, max_tokens=1000)
-    answer = response.choices[0].message.content
-    return answer
-```
-
-Next, initialize an engine to load a model. To run an agent locally, create a [`TransformersEngine`] to load a preinitialized [`Pipeline`].
-
-However, you could also leverage Hugging Face's powerful inference infrastructure, [Inference API](https://hf.co/docs/api-inference/index) or [Inference Endpoints](https://hf.co/docs/inference-endpoints/index), to run your model. This is useful for loading larger models that are typically required for agentic behavior. In this case, load the [`HfApiEngine`] to run the agent.
-
-The agent requires a list of tools it can use to complete a task. If you aren't using any additional tools, pass an empty list. The default tools provided by Transformers are loaded automatically, but you can optionally set `add_base_tools=True` to explicitly enable them.
-
-<hfoptions id="engine">
-<hfoption id="TransformersEngine">
-
-```py
-from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TransformersEngine, CodeAgent
-
-tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-3.1-8B-Instruct")
-model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-3.1-8B-Instruct").to("cuda")
-pipeline = pipeline("text-generation", model=model, tokenizer=tokenizer)
-llm_engine = TransformersEngine(pipeline)
-agent = CodeAgent(tools=[], llm_engine=llm_engine)
-agent.run(
-    "What causes bread to rise?",
-)
-```
-
-</hfoption>
-<hfoption id="HfApiEngine">
-
-```py
-from transformers import CodeAgent, HfApiEngine
-
-llm_engine = HfApiEngine(model="meta-llama/Meta-Llama-3-70B-Instruct")
-agent = CodeAgent(tools=[], llm_engine=llm_engine)
-agent.run(
-    "Could you translate this sentence from French, say it out loud and return the audio.",
-    sentence="Où est la boulangerie la plus proche?",
-)
-```
-
-</hfoption>
-</hfoptions>
-
-The agent supports [constrained generation](https://hf.co/docs/text-generation-inference/conceptual/guidance) for generating outputs according to a specific structure with the `grammar` parameter. The `grammar` parameter should be specified in the `llm_engine` method or you can set it when initializing an agent.
-
-Lastly, an agent accepts additional inputs such as text and audio. In the [`HfApiEngine`] example above, the agent accepted a sentence to translate. But you could also pass a path to a local or remote file for the agent to access. The example below demonstrates how to pass a path to an audio file.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], llm_engine=llm_engine)
-agent.run("Why doesn't he know many people in New York?", audio="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/recording.mp3")
-```
-
-## System prompt
-
-A system prompt describes how an agent should behave, a description of the available tools, and the expected output format.
-
-Tools are defined by the `<<tool_descriptions>>` token which is dynamically replaced during runtime with the actual tool. The tool description is derived from the tool name, description, inputs, output type, and a Jinja2 template. Refer to the [Tools](./tools) guide for more information about how to describe tools.
-
-The example below is the system prompt for [`ReactCodeAgent`].
-
-```py
-You will be given a task to solve as best you can.
-You have access to the following tools:
-<<tool_descriptions>>
-
-To solve the task, you must plan forward to proceed in a series of steps, in a cycle of 'Thought:', 'Code:', and 'Observation:' sequences.
-
-At each step, in the 'Thought:' sequence, you should first explain your reasoning towards solving the task, then the tools that you want to use.
-Then in the 'Code:' sequence, you should write the code in simple Python. The code sequence must end with '/End code' sequence.
-During each intermediate step, you can use 'print()' to save whatever important information you will then need.
-These print outputs will then be available in the 'Observation:' field, for using this information as input for the next step.
-
-In the end you have to return a final answer using the `final_answer` tool.
-
-Here are a few examples using notional tools:
----
-{examples}
-
-Above example were using notional tools that might not exist for you. You only have access to those tools:
-<<tool_names>>
-You also can perform computations in the python code you generate.
-
-Always provide a 'Thought:' and a 'Code:\n```py' sequence ending with '```<end_code>' sequence. You MUST provide at least the 'Code:' sequence to move forward.
-
-Remember to not perform too many operations in a single code block! You should split the task into intermediate code blocks.
-Print results at the end of each step to save the intermediate results. Then use final_answer() to return the final result.
-
-Remember to make sure that variables you use are all defined.
-
-Now Begin!
-```
-
-The system prompt can be tailored to the intended task. For example, you can add a better explanation of the output format or you can overwrite the system prompt template entirely with your own custom system prompt as shown below.
-
-> [!WARNING]
-> If you're writing a custom system prompt, make sure to include `<<tool_descriptions>>` in the template so the agent is aware of the available tools.
-
-```py
-from transformers import ReactJsonAgent
-from transformers.agents import PythonInterpreterTool
-
-agent = ReactJsonAgent(tools=[PythonInterpreterTool()], system_prompt="{your_custom_prompt}")
-```
-
-## Code execution
-
-For safety, only the tools you provide (and the default Transformers tools) and the `print` function are executed. The interpreter doesn't allow importing modules that aren't on a safe list.
-
-To import modules that aren't on the list, add them as a list to the `additional_authorized_imports` parameter when initializing an agent.
-
-```py
-from transformers import ReactCodeAgent
-
-agent = ReactCodeAgent(tools=[], additional_authorized_imports=['requests', 'bs4'])
-agent.run("Could you get me the title of the page at url 'https://huggingface.co/blog'?")
-```
-
-Code execution stops if a tool isn't on the safe list, it isn't authorized, or if the code generated by the agent returns a Python error.
-
-> [!WARNING]
-> A LLM can generate any arbitrary code that can be executed, so don't add any unsafe imports!
-
-## Multi-agent
-
-[Multi-agent](https://hf.co/papers/2308.08155) refers to multiple agents working together to solve a task. Performance is typically better because each agent is specialized for a particular subtask.
-
-Multi-agents are created through a [`ManagedAgent`] class, where a *manager agent* oversees how other agents work together. The manager agent requires an agent and their name and description. These are added to the manager agents system prompt which lets it know how to call and use them.
-
-The multi-agent example below creates a web search agent that is managed by another [`ReactCodeAgent`].
-
-```py
-from transformers.agents import ReactCodeAgent, HfApiEngine, DuckDuckGoSearchTool, ManagedAgent
-
-llm_engine = HfApiEngine()
-web_agent = ReactCodeAgent(tools=[DuckDuckGoSearchTool()], llm_engine=llm_engine)
-managed_web_agent = ManagedAgent(
-    agent=web_agent,
-    name="web_search",
-    description="Runs web searches for you. Give it your query as an argument."
-)
-manager_agent = ReactCodeAgent(
-    tools=[], llm_engine=llm_engine, managed_agents=[managed_web_agent]
-)
-manager_agent.run("Who is the CEO of Hugging Face?")
-```
-
-## Gradio integration
-
-[Gradio](https://www.gradio.app/) is a library for quickly creating and sharing machine learning apps. The [gradio.Chatbot](https://www.gradio.app/docs/gradio/chatbot) supports chatting with a Transformers agent with the [`stream_to_gradio`] function.
-
-Load a tool and LLM with an agent, and then create a Gradio app. The key is to use [`stream_to_gradio`] to stream the agents messages and display how it's reasoning through a task.
-
-```py
-import gradio as gr
-from transformers import (
-    load_tool,
-    ReactCodeAgent,
-    HfApiEngine,
-    stream_to_gradio,
-)
-
-# Import tool from Hub
-image_generation_tool = load_tool("m-ric/text-to-image")
-llm_engine = HfApiEngine("meta-llama/Meta-Llama-3-70B-Instruct")
-
-# Initialize the agent with the image generation tool
-agent = ReactCodeAgent(tools=[image_generation_tool], llm_engine=llm_engine)
-
-def interact_with_agent(task):
-    messages = []
-    messages.append(gr.ChatMessage(role="user", content=task))
-    yield messages
-    for msg in stream_to_gradio(agent, task):
-        messages.append(msg)
-        yield messages + [
-            gr.ChatMessage(role="assistant", content="⏳ Task not finished yet!")
-        ]
-    yield messages
-
-with gr.Blocks() as demo:
-    text_input = gr.Textbox(lines=1, label="Chat Message", value="Make me a picture of the Statue of Liberty.")
-    submit = gr.Button("Run illustrator agent!")
-    chatbot = gr.Chatbot(
-        label="Agent",
-        type="messages",
-        avatar_images=(
-            None,
-            "https://em-content.zobj.net/source/twitter/53/robot-face_1f916.png",
-        ),
-    )
-    submit.click(interact_with_agent, [text_input], [chatbot])
-
-if __name__ == "__main__":
-    demo.launch()
-```
-
-## Troubleshoot
-
-For a better idea of what is happening when you call an agent, it is always a good idea to check the system prompt template first.
-
-```py
-print(agent.system_prompt_template)
-```
-
-If the agent is behaving unexpectedly, remember to explain the task you want to perform as clearly as possible. Every [`~Agent.run`] is different and minor variations in your system prompt may yield completely different results.
-
-To find out what happened after a run, check the following agent attributes.
-
-- `agent.logs` stores the finegrained agent logs. At every step of the agents run, everything is stored in a dictionary and appended to `agent.logs`.
-- `agent.write_inner_memory_from_logs` only stores a high-level overview of the agents run. For example, at each step, it stores the LLM output as a message and the tool call output as a separate message. Not every detail from a step is transcripted by `write_inner_memory_from_logs`.
-
-## Resources
-
-Learn more about ReAct agents in the [Open-source LLMs as LangChain Agents](https://hf.co/blog/open-source-llms-as-agents) blog post.
+> Agents and tools were spun out into the standalone [smolagents](https://huggingface.co/docs/smolagents/index) library. They were removed from `transformers` in v4.52.

docs/source/en/attention_interface.md

@@ -0,0 +1,128 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Attention Interface
+
+This page describes how to use the `AttentionInterface` in order to register custom attention functions to use with
+supported models.
+
+## Customizing attention function
+
+Most recent models can now switch from one attention function used in the Attention layer to the other, thanks to a simple mapping.
+By default, we provide the implementation for [`sdpa`](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html),
+[`flash_attention_2`](https://github.com/Dao-AILab/flash-attention) and [`flex_attention`](https://pytorch.org/docs/stable/nn.attention.flex_attention.html#module-torch.nn.attention.flex_attention)
+as well as `eager`, which is a simple matrix multiplication without any optimization on top.  
+This is the setting you can usually choose when instantiating a model:
+
+```python
+from transformers import AutoModelForCausalLM
+
+model_id = "meta-llama/Llama-3.2-1B"
+
+# Here, using flash attention as an example
+model = AutoModelForCausalLM.from_pretrained(model_id, attn_implementation="flash_attention_2")
+```
+
+But what if you wanted to create your own attention function? Or simply play around with existing ones, adding
+a few statements here and there? You can now do so with the `AttentionInterface`! Here is an example:
+
+```python
+from transformers import AutoModelForCausalLM, AttentionInterface
+from transformers.integrations.sdpa_attention import sdpa_attention_forward
+import torch
+
+model_id = "meta-llama/Llama-3.2-1B"
+
+def my_new_sdpa(*args, **kwargs):
+    print("I just entered the attention computation")
+    return sdpa_attention_forward(*args, **kwargs)
+
+AttentionInterface.register("my_new_sdpa", my_new_sdpa)
+
+model = AutoModelForCausalLM.from_pretrained(model_id, attn_implementation="my_new_sdpa")
+# Try running the forward with the new attention function
+model(torch.ones(1, 5, dtype=int))
+```
+
+You will see it prints "I just entered the attention computation" as many times as there are layers in the model (with this example, 16 times).
+
+## Dynamically switching attention function
+
+You could dynamically change the model's attention function as well, by overriding the `config._attn_implementation` field:
+
+```python
+# Back to use original sdpa implementation
+model.config._attn_implementation = "sdpa"
+
+model(torch.ones(1, 5, dtype=int))
+```
+
+and it will stop printing the statements, as it now uses the `sdpa` attention.  
+This allows to quickly change an attention function, without needing to reload the model!
+
+## What about new args needed in my custom attention function?
+
+But indeed, what if the new function requires a new arg to be properly used? It's no issue! Models supporting the
+`AttentionInterface` propagate kwargs all the way to the Attention layers, and to the used attention function. That way,
+you can simply pass the arg (as a kwargs, i.e. you need to qualify the name of the arg) in the model's forward, and it will be correctly used in the attention. However, custom attention functions have some limitations. In particular, it must follow the signature and return format of other attention functions, i.e.
+
+```python
+from transformers import AutoModelForCausalLM, AttentionInterface
+from transformers.integrations.sdpa_attention import sdpa_attention_forward
+import torch
+
+def custom_attention(
+    module: torch.nn.Module,  # required arg
+    query: torch.Tensor,  # required arg
+    key: torch.Tensor,  # required arg
+    value: torch.Tensor,  # required arg
+    attention_mask: Optional[torch.Tensor],  # required arg
+    a_new_kwargs = None,  # You can now add as many kwargs as you need
+    another_new_kwargs = None,  # You can now add as many kwargs as you need
+    **kwargs,  # You need to accept **kwargs as models will pass other args
+) -> Tuple[torch.Tensor, Optional[torch.Tensor]]
+    ...  # do your magic!
+    return attn_output, attn_weights  # attn_weights are optional here
+
+AttentionInterface.register("custom", custom_attention)
+
+model = AutoModelForCausalLM.from_pretrained(model_id, attn_implementation="custom")
+# Forward pass with the new kwargs
+model(torch.ones(1, 5, dtype=int), a_new_kwargs=..., another_new_kwargs=...)
+```
+
+If in doubt about what args/kwargs a given model sends to the attention function, simply check that model's modeling code on [GitHub](https://github.com/huggingface/transformers/tree/main/src/transformers/models)!
+
+## Accessing current available implementations
+
+Most of the time, you will simply need to `register` a new function. If, however, you need to access an existing one,
+and/or perform a few checks, the prefered way is to use the global `ALL_ATTENTION_FUNCTIONS`. It behaves the same way you
+would expect from a usual Python dictionary:
+
+```python
+>>> from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS
+
+>>> list(ALL_ATTENTION_FUNCTIONS.keys())
+>>> ['flash_attention_2', 'flex_attention', 'sdpa']
+
+>>> ALL_ATTENTION_FUNCTIONS["sdpa"]
+>>> <function transformers.integrations.sdpa_attention.sdpa_attention_forward>
+
+>>> ALL_ATTENTION_FUNCTIONS.get("sdpa", None)
+>>> <function transformers.integrations.sdpa_attention.sdpa_attention_forward>
+
+# You can also globally `register` a new function directly on it
+>>> ALL_ATTENTION_FUNCTIONS.register("new_func", new_func)
+```

docs/source/en/cache_explanation.md

@@ -9,7 +9,7 @@ Unless required by applicable law or agreed to in writing, software distributed
 an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
 specific language governing permissions and limitations under the License.
 
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
 
 -->
@@ -62,7 +62,7 @@ for _ in range(max_new_tokens):
     # Greedily sample one next token
     next_token_ids = outputs.logits[:, -1:].argmax(-1)
     generated_ids = torch.cat([generated_ids, next_token_ids], dim=-1)
-    # Prepare inputs for the next generation step by leaaving unprocessed tokens, in our case we have only one new token
+    # Prepare inputs for the next generation step by leaving unprocessed tokens, in our case we have only one new token
     # and expanding attn mask for the new token, as explained above
     attention_mask = inputs["attention_mask"]
     attention_mask = torch.cat([attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1)
@@ -88,7 +88,7 @@ model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-chat-hf", to
 inputs = tokenizer("Hello, my name is", return_tensors="pt").to(model.device)
 
 # `return_dict_in_generate=True` is required to return the cache and `return_legacy_cache` forces the returned cache
-# in the the legacy format
+# in the legacy format
 generation_outputs = model.generate(**inputs, return_dict_in_generate=True, return_legacy_cache=True, max_new_tokens=5)
 
 cache = DynamicCache.from_legacy_cache(generation_outputs.past_key_values)

docs/source/en/chat_extras.md

@@ -146,7 +146,7 @@ print(tokenizer.decode(out[0][len(inputs["input_ids"][0]):]))
 
 ## Schema
 
-[`~PreTrainedTokenizerBase.apply_chat_template`] converts functions into a [JSON schema](https://json-schema.org/learn/getting-started-step-by-step) which is passed to the chat template. A LLM never sees the code inside the function. In other words, a LLM doesn't care how the model works technically, it only cares about function **definition** and **arguments**.
+[`~PreTrainedTokenizerBase.apply_chat_template`] converts functions into a [JSON schema](https://json-schema.org/learn/getting-started-step-by-step) which is passed to the chat template. A LLM never sees the code inside the function. In other words, a LLM doesn't care how the function works technically, it only cares about function **definition** and **arguments**.
 
 The JSON schema is automatically generated behind the scenes as long as your function follows the [rules](#tools) listed earlier above. But you can use [get_json_schema](https://github.com/huggingface/transformers/blob/14561209291255e51c55260306c7d00c159381a5/src/transformers/utils/chat_template_utils.py#L205) to manually convert a schema for more visibility or debugging.
 

docs/source/en/chat_templating_multimodal.md

@@ -9,7 +9,7 @@ Unless required by applicable law or agreed to in writing, software distributed
 an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
 specific language governing permissions and limitations under the License.
 
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
 
 -->
@@ -18,7 +18,7 @@ rendered properly in your Markdown viewer.
 
 Multimodal model chat templates expect a similar [template](./chat_templating) as text-only models. It needs `messages` that includes a dictionary of the `role` and `content`.
 
-Multimodal templates are included in the [Processor](./processors) class and requires an additional `type` key for specifying whether the included content is an image, video, or text.
+Multimodal templates are included in the [Processor](./processors) class and require an additional `type` key for specifying whether the included content is an image, video, or text.
 
 This guide will show you how to format chat templates for multimodal models as well as some best practices for configuring the template
 
@@ -109,7 +109,7 @@ These inputs are now ready to be used in [`~GenerationMixin.generate`].
 
 Some vision models also support video inputs. The message format is very similar to the format for [image inputs](#image-inputs).
 
-- The content `"type"` should be `"video"` to indicate the the content is a video.
+- The content `"type"` should be `"video"` to indicate the content is a video.
 - For videos, it can be a link to the video (`"url"`) or it could be a file path (`"path"`). Videos loaded from a URL can only be decoded with [PyAV](https://pyav.basswood-io.com/docs/stable/) or [Decord](https://github.com/dmlc/decord).
 
 > [!WARNING]
@@ -141,7 +141,7 @@ Pass `messages` to [`~ProcessorMixin.apply_chat_template`] to tokenize the input
 
 The `video_load_backend` parameter refers to a specific framework to load a video. It supports [PyAV](https://pyav.basswood-io.com/docs/stable/), [Decord](https://github.com/dmlc/decord), [OpenCV](https://github.com/opencv/opencv), and [torchvision](https://pytorch.org/vision/stable/index.html).
 
-The examples below uses Decord as the backend because it is a bit faster than PyAV.
+The examples below use Decord as the backend because it is a bit faster than PyAV.
 
 <hfoptions id="sampling">
 <hfoption id="fixed number of frames">
@@ -181,35 +181,6 @@ processed_chat = processor.apply_chat_template(
 print(processed_chat.keys())
 ```
 
-</hfoption>
-<hfoption id="custom frame sampling">
-
-Some models don't sample frames *uniformly* and require more complex logic to determine which frames to use. For example, the model may have an *adaptive frame selection* or if the model prioritizes *key moments* in a video rather than evenly spaced frames.
-
-If a model has a different sampling strategy, you can write a function that customizes frame selection. The function should include the following requirements.
-
-- Use the `sample_indices_fn` parameter to pass a callable function for sampling.
-- If provided, this function *overrides* the standard `num_frames` and `fps` parameters.
-- The function receives all the parameters passed to `load_video` and must return valid frame indices to sample from.
-
-An example function is shown below. This gives you full control over frame selection, making the model more adaptable to different video scenarios.
-
-```py
-def sample_indices_fn(metadata, **kwargs):
-    # samples only the first and the second frame
-    return [0, 1]
-
-processed_chat = processor.apply_chat_template(
-    messages,
-    add_generation_prompt=True,
-    tokenize=True,
-    return_dict=True,
-    sample_indices_fn=sample_indices_fn,
-    video_load_backend="decord",
-)
-print(processed_chat.keys())
-```
-
 </hfoption>
 <hfoption id="list of image frames">
 

docs/source/en/custom_models.md

@@ -131,7 +131,7 @@ class ResnetModel(PreTrainedModel):
 </hfoption>
 <hfoption id="ResnetModelForImageClassification">
 
-The `forward` method needs to be rewrittten to calculate the loss for each logit if labels are available. Otherwise, the ResNet model class is the same.
+The `forward` method needs to be rewritten to calculate the loss for each logit if labels are available. Otherwise, the ResNet model class is the same.
 
 > [!TIP]
 > Add `config_class` to the model class to enable [AutoClass](#autoclass-support) support.

docs/source/en/generation_strategies.md

@@ -31,7 +31,7 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
-inputs = tokenizer("I look forward to", return_tensors="pt").to("cuda")
+inputs = tokenizer("Hugging Face is an open-source company", return_tensors="pt").to("cuda")
 
 model = AutoModelForCausalLM.from_pretrained("meta-llama/Llama-2-7b-hf", torch_dtype=torch.float16).to("cuda")
 # explicitly set to default length because Llama2 generation length is 4096
@@ -271,7 +271,7 @@ tokenizer.batch_decode(outputs, skip_special_tokens=True)
 
 ## DoLa
 
-[Decoding by Contrasting Layers (DoLa)](https://hf.co/papers/2309.03883) is a contrastive decoding strategy for improving factuality and reducing hallucination. This strategy works by contrasting the logit diffferences between the final and early layers. As a result, factual knowledge localized to particular layers are amplified. DoLa is not recommended for smaller models like GPT-2.
+[Decoding by Contrasting Layers (DoLa)](https://hf.co/papers/2309.03883) is a contrastive decoding strategy for improving factuality and reducing hallucination. This strategy works by contrasting the logit differences between the final and early layers. As a result, factual knowledge localized to particular layers are amplified. DoLa is not recommended for smaller models like GPT-2.
 
 Enable DoLa with the following parameters.
 

docs/source/en/gguf.md

@@ -24,21 +24,23 @@ rendered properly in your Markdown viewer.
 
 The GGUF format also supports many quantized data types (refer to [quantization type table](https://hf.co/docs/hub/en/gguf#quantization-types) for a complete list of supported quantization types) which saves a significant amount of memory, making inference with large models like Whisper and Llama feasible on local and edge devices.
 
-Transformers supports loading models stored in the GGUF format for further training or finetuning. The GGUF format is dequantized to fp32 where the full model weights are available and compatible with PyTorch.
+Transformers supports loading models stored in the GGUF format for further training or finetuning. The GGUF checkpoint is **dequantized to fp32** where the full model weights are available and compatible with PyTorch.
 
 > [!TIP]
-> Models that support GGUF include Llama, Mistral, Qwen2, Qwen2Moe, Phi3, Bloom, Falcon, StableLM, GPT2, and Starcoder2.
+> Models that support GGUF include Llama, Mistral, Qwen2, Qwen2Moe, Phi3, Bloom, Falcon, StableLM, GPT2, Starcoder2, and [more](https://github.com/huggingface/transformers/blob/main/src/transformers/integrations/ggml.py)
 
 Add the `gguf_file` parameter to [`~PreTrainedModel.from_pretrained`] to specify the GGUF file to load.
 
 ```py
+# pip install gguf
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
 model_id = "TheBloke/TinyLlama-1.1B-Chat-v1.0-GGUF"
 filename = "tinyllama-1.1b-chat-v1.0.Q6_K.gguf"
 
+torch_dtype = torch.float32 # could be torch.float16 or torch.bfloat16 too
 tokenizer = AutoTokenizer.from_pretrained(model_id, gguf_file=filename)
-model = AutoModelForCausalLM.from_pretrained(model_id, gguf_file=filename)
+model = AutoModelForCausalLM.from_pretrained(model_id, gguf_file=filename, torch_dtype=torch_dtype)
 ```
 
 Once you're done tinkering with the model, save and convert it back to the GGUF format with the [convert-hf-to-gguf.py](https://github.com/ggerganov/llama.cpp/blob/master/convert_hf_to_gguf.py) script.

docs/source/en/gpu_selection.md

@@ -9,7 +9,7 @@ Unless required by applicable law or agreed to in writing, software distributed
 an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
 specific language governing permissions and limitations under the License.
 
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
 
 -->
@@ -56,7 +56,7 @@ deepspeed --num_gpus 2 trainer-program.py ...
 
 ### Order of GPUs
 
-To select specific GPUs to use and their order, configure the the `CUDA_VISIBLE_DEVICES` environment variable. It is easiest to set the environment variable in `~/bashrc` or another startup config file. `CUDA_VISIBLE_DEVICES` is used to map which GPUs are used. For example, if there are 4 GPUs (0, 1, 2, 3) and you only want to run GPUs 0 and 2:
+To select specific GPUs to use and their order, configure the `CUDA_VISIBLE_DEVICES` environment variable. It is easiest to set the environment variable in `~/bashrc` or another startup config file. `CUDA_VISIBLE_DEVICES` is used to map which GPUs are used. For example, if there are 4 GPUs (0, 1, 2, 3) and you only want to run GPUs 0 and 2:
 
 ```bash
 CUDA_VISIBLE_DEVICES=0,2 torchrun trainer-program.py ...

docs/source/en/how_to_hack_models.md

@@ -36,7 +36,7 @@ This guide will show you how to customize a models attention mechanism in order
 
 ## Attention class
 
-[Segment Anything](./model_doc/sam) is an image segmentation model, and it combines the query-key-value (`qkv`) projection in its attention mechanims. To reduce the number of trainable parameters and computational overhead, you can apply LoRA to the `qkv` projection. This requires splitting the `qkv` projection so that you can separately target the `q` and `v` with LoRA.
+[Segment Anything](./model_doc/sam) is an image segmentation model, and it combines the query-key-value (`qkv`) projection in its attention mechanisms. To reduce the number of trainable parameters and computational overhead, you can apply LoRA to the `qkv` projection. This requires splitting the `qkv` projection so that you can separately target the `q` and `v` with LoRA.
 
 1. Create a custom attention class, `SamVisionAttentionSplit`, by subclassing the original `SamVisionAttention` class. In the `__init__`, delete the combined `qkv` and create a separate linear layer for `q`, `k` and `v`.
 

docs/source/en/index.md

@@ -43,4 +43,3 @@ Transformers is designed for developers and machine learning engineers and resea
   </a>
 </div>
 
-Join us on the Hugging Face [Hub](https://huggingface.co/), [Discord](https://discord.com/invite/JfAtkvEtRb), or [forum](https://discuss.huggingface.co/) to collaborate and build models, datasets, and applications together.

docs/source/en/installation.md

@@ -20,7 +20,7 @@ rendered properly in your Markdown viewer.
 
 # Installation
 
-Transformers works with [PyTorch](https://pytorch.org/get-started/locally/), [TensorFlow 2.0](https://www.tensorflow.org/install/pip), and [Flax](https://flax.readthedocs.io/en/latest/). It has been tested on Python 3.6+, PyTorch 1.1.0+, TensorFlow 2.0+, and Flax.
+Transformers works with [PyTorch](https://pytorch.org/get-started/locally/), [TensorFlow 2.0](https://www.tensorflow.org/install/pip), and [Flax](https://flax.readthedocs.io/en/latest/). It has been tested on Python 3.9+, PyTorch 2.1+, TensorFlow 2.6+, and Flax 0.4.1+.
 
 ## Virtual environment
 
@@ -33,7 +33,7 @@ Create and activate a virtual environment in your project directory with [venv](
 
 ```bash
 python -m venv .env
-source ./env/bin/activate
+source .env/bin/activate
 ```
 
 </hfoption>
@@ -43,7 +43,7 @@ source ./env/bin/activate
 
 ```bash
 uv venv .env
-source ./env/bin/activate
+source .env/bin/activate
 ```
 
 </hfoption>

docs/source/en/internal/import_utils.md

@@ -0,0 +1,91 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Import Utilities
+
+This page goes through the transformers utilities to enable lazy and fast object import.
+While we strive for minimal dependencies, some models have specific dependencies requirements that cannot be
+worked around. We don't want for all users of `transformers` to have to install those dependencies to use other models,
+we therefore mark those as soft dependencies rather than hard dependencies.
+
+The transformers toolkit is not made to error-out on import of a model that has a specific dependency; instead, an
+object for which you are lacking a dependency will error-out when calling any method on it. As an example, if 
+`torchvision` isn't installed, the fast image processors will not be available. 
+
+This object is still importable:
+
+```python
+>>> from transformers import DetrImageProcessorFast
+>>> print(DetrImageProcessorFast)
+<class 'DetrImageProcessorFast'>
+```
+
+However, no method can be called on that object:
+
+```python
+>>> DetrImageProcessorFast.from_pretrained()
+ImportError: 
+DetrImageProcessorFast requires the Torchvision library but it was not found in your environment. Checkout the instructions on the
+installation page: https://pytorch.org/get-started/locally/ and follow the ones that match your environment.
+Please note that you may need to restart your runtime after installation.
+```
+
+Let's see how to specify specific object dependencies.
+
+## Specifying Object Dependencies
+
+### Filename-based
+
+All objects under a given filename have an automatic dependency to the tool linked to the filename
+
+**TensorFlow**: All files starting with `modeling_tf_` have an automatic TensorFlow dependency.
+
+**Flax**: All files starting with `modeling_flax_` have an automatic Flax dependency
+
+**PyTorch**: All files starting with `modeling_` and not valid with the above (TensorFlow and Flax) have an automatic 
+PyTorch dependency
+
+**Tokenizers**: All files starting with `tokenization_` and ending with `_fast` have an automatic `tokenizers` dependency
+
+**Vision**: All files starting with `image_processing_` have an automatic dependency to the `vision` dependency group; 
+at the time of writing, this only contains the `pillow` dependency.
+
+**Vision + Torch + Torchvision**: All files starting with `image_processing_` and ending with `_fast` have an automatic
+dependency to `vision`, `torch`, and `torchvision`.
+
+All of these automatic dependencies are added on top of the explicit dependencies that are detailed below.
+
+### Explicit Object Dependencies
+
+We add a method called `requires` that is used to explicitly specify the dependencies of a given object. As an
+example, the `Trainer` class has two hard dependencies: `torch` and `accelerate`. Here is how we specify these 
+required dependencies:
+
+```python
+from .utils.import_utils import requires
+
+@requires(backends=("torch", "accelerate"))
+class Trainer:
+    ...
+```
+
+Backends that can be added here are all the backends that are available in the `import_utils.py` module.
+
+## Methods
+
+[[autodoc]] utils.import_utils.define_import_structure
+
+[[autodoc]] utils.import_utils.requires

docs/source/en/internal/model_debugging_utils.md

@@ -0,0 +1,213 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# Model debugging toolboxes
+
+This page lists all the debugging and model adding tools used by the library, as well as the utility functions it provides for it.
+
+Most of those are only useful if you are adding new models in the library.
+
+
+## Model addition debuggers
+
+
+### Model addition debugger - context manager for model adders
+
+This context manager is a power user tool intended for model adders.
+It tracks all forward calls within a model forward and logs a slice of each input and output on a nested Json.
+To note, this context manager enforces `torch.no_grad()`.
+
+### Rationale
+
+Because when porting models to transformers, even from python to python, model adders often have to do a lot of manual operations, involving saving and loading tensors, comparing dtypes, etc. This small tool can hopefully shave off some time.
+
+### Usage
+
+Add this context manager as follows to debug a model:
+
+```python
+import torch
+from PIL import Image
+import requests
+from transformers import LlavaProcessor, LlavaForConditionalGeneration
+from transformers.model_debugging_utils import model_addition_debugger_context
+torch.random.manual_seed(673)
+
+# load pretrained model and processor
+model_id = "llava-hf/llava-1.5-7b-hf"
+processor = LlavaProcessor.from_pretrained(model_id)
+model = LlavaForConditionalGeneration.from_pretrained(model_id, low_cpu_mem_usage=True)
+
+# create random image input
+random_image = Image.fromarray(torch.randint(0, 256, (224, 224, 3), dtype=torch.uint8).numpy())
+
+# prompt
+prompt = "<image>Describe this image."
+
+# process inputs
+inputs = processor(text=prompt, images=random_image, return_tensors="pt")
+
+# call forward method (not .generate!)
+with model_addition_debugger_context(
+  model,
+  debug_path="optional_path_to_your_directory",
+  do_prune_layers=False # This will output ALL the layers of a model.
+  ):
+    output = model.forward(**inputs)
+
+```
+
+
+### Reading results
+
+The debugger generates two files from the forward call, both with the same base name, 
+but ending either with `_SUMMARY.json` or with `_FULL_TENSORS.json`. 
+
+The first one will contain a summary of each module's _input_ and _output_ tensor values and shapes.
+
+```json
+{
+  "module_path": "MolmoForConditionalGeneration",
+  "inputs": {
+    "args": [],
+    "kwargs": {
+      "input_ids": {
+        "shape": "torch.Size([1, 589])",
+        "dtype": "torch.int64"
+      },
+      "attention_mask": {
+        "shape": "torch.Size([1, 589])",
+        "dtype": "torch.int64"
+      },
+      "pixel_values": {
+        "shape": "torch.Size([1, 5, 576, 588])",
+        "dtype": "torch.float32",
+        "mean": "tensor(-8.9514e-01, device='cuda:0')",
+        "std": "tensor(9.2586e-01, device='cuda:0')",
+        "min": "tensor(-1.7923e+00, device='cuda:0')",
+        "max": "tensor(1.8899e+00, device='cuda:0')"
+    }
+  },
+  "children": [
+    {
+      "module_path": "MolmoForConditionalGeneration.language_model.model.embed_tokens",
+      "inputs": {
+        "args": [
+          {
+            "shape": "torch.Size([1, 589])",
+            "dtype": "torch.int64"
+          }
+        ]
+      },
+      "outputs": {
+        "shape": "torch.Size([1, 589, 3584])",
+        "dtype": "torch.float32",
+        "mean": "tensor(6.5460e-06, device='cuda:0')",
+        "std": "tensor(2.3807e-02, device='cuda:0')",
+        "min": "tensor(-3.3398e-01, device='cuda:0')",
+        "max": "tensor(3.9453e-01, device='cuda:0')"
+      }
+    },
+    {
+      "module_path": "MolmoForConditionalGeneration.vision_tower",
+      "inputs": {
+        "args": [
+          {
+            "shape": "torch.Size([5, 1, 576, 588])",
+            "dtype": "torch.float32",
+            "mean": "tensor(-8.9514e-01, device='cuda:0')",
+            "std": "tensor(9.2586e-01, device='cuda:0')",
+            "min": "tensor(-1.7923e+00, device='cuda:0')",
+            "max": "tensor(1.8899e+00, device='cuda:0')"
+          }
+        ],
+        "kwargs": {
+          "output_hidden_states": "True"
+        }
+      },
+      "children": [
+        { ... and so on
+```
+
+The `_FULL_TENSORS.json` file will display a full view of all tensors, which is useful
+for comparing two files. 
+```json
+      "pixel_values": {
+        "shape": "torch.Size([1, 5, 576, 588])",
+        "dtype": "torch.float32",
+        "value": [
+          "tensor([[[[-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          ...,",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00]],",
+          "",
+          "         [[-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          ...,",
+          "          [-1.4857e+00, -1.4820e+00, -1.2100e+00,  ..., -6.0979e-01, -5.9650e-01, -3.8527e-01],",
+          "          [-1.6755e+00, -1.7221e+00, -1.4518e+00,  ..., -7.5577e-01, -7.4658e-01, -5.5592e-01],",
+          "          [-7.9957e-01, -8.2162e-01, -5.7014e-01,  ..., -1.3689e+00, -1.3169e+00, -1.0678e+00]],",
+          "",
+          "         [[-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          ...,",
+          "          [-3.0322e-01, -5.0645e-01, -5.8436e-01,  ..., -6.2439e-01, -7.9160e-01, -8.1188e-01],",
+          "          [-4.4921e-01, -6.5653e-01, -7.2656e-01,  ..., -3.4702e-01, -5.2146e-01, -5.1326e-01],",
+          "          [-3.4702e-01, -5.3647e-01, -5.4170e-01,  ..., -1.0915e+00, -1.1968e+00, -1.0252e+00]],",
+          "",
+          "         [[-1.1207e+00, -1.2718e+00, -1.0678e+00,  ..., 1.2013e-01, -1.3126e-01, -1.7197e-01],",
+          "          [-6.9738e-01, -9.1166e-01, -8.5454e-01,  ..., -5.5050e-02, -2.8134e-01, -4.2793e-01],",
+          "          [-3.4702e-01, -5.5148e-01, -5.8436e-01,  ..., 1.9312e-01, -8.6235e-02, -2.1463e-01],",
+          "          ...,",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00]],",
+          "",
+          "         [[-1.0039e+00, -9.5669e-01, -6.5546e-01,  ..., -1.4711e+00, -1.4219e+00, -1.1389e+00],",
+          "          [-1.0039e+00, -9.5669e-01, -6.5546e-01,  ..., -1.7193e+00, -1.6771e+00, -1.4091e+00],",
+          "          [-1.6317e+00, -1.6020e+00, -1.2669e+00,  ..., -1.2667e+00, -1.2268e+00, -8.9720e-01],",
+          "          ...,",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00],",
+          "          [-1.7923e+00, -1.7521e+00, -1.4802e+00,  ..., -1.7923e+00, -1.7521e+00, -1.4802e+00]]]], device='cuda:0')"
+        ],
+        "mean": "tensor(-8.9514e-01, device='cuda:0')",
+        "std": "tensor(9.2586e-01, device='cuda:0')",
+        "min": "tensor(-1.7923e+00, device='cuda:0')",
+        "max": "tensor(1.8899e+00, device='cuda:0')"
+      },
+```
+
+### Comparing between implementations
+
+Once the forward passes of two models have been traced by the debugger, one can compare the `json` output files. See below: we can see slight differences between these two implementations' key projection layer. Inputs are mostly identical, but not quite. Looking through the file differences makes it easier to pinpoint which layer is wrong. 
+
+
+![download-icon](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/files_difference_debugging.png)
+
+
+### Limitations and scope
+
+This feature will only work for torch-based models, and would require more work and case-by-case approach for say `jax`-based models that are usually compiled. Models relying heavily on external kernel calls may work, but trace will probably miss some things. Regardless, any python implementation that aims at mimicking another implementation can be traced once instead of reran N times with breakpoints.
+
+If you pass `do_prune_layers=False` to your model debugger, ALL the layers will be outputted to `json`. Else, only the first and last layer will be shown. This is useful when some layers (typically cross-attention) appear only after N layers. 
+
+[[autodoc]] model_addition_debugger_context

docs/source/en/internal/modeling_utils.md

@@ -16,32 +16,27 @@ rendered properly in your Markdown viewer.
 
 # Custom Layers and Utilities
 
-This page lists all the custom layers used by the library, as well as the utility functions it provides for modeling.
+This page lists all the custom layers used by the library, as well as the utility functions and classes it provides for modeling.
 
 Most of those are only useful if you are studying the code of the models in the library.
 
+## Layers
+
+[[autodoc]] GradientCheckpointingLayer
+
+## Attention Functions
+
+[[autodoc]] AttentionInterface
+    - register
+
+## Rotary Position Embedding Functions
+
+[[autodoc]] dynamic_rope_update
 
 ## Pytorch custom modules
 
 [[autodoc]] pytorch_utils.Conv1D
 
-[[autodoc]] modeling_utils.PoolerStartLogits
-    - forward
-
-[[autodoc]] modeling_utils.PoolerEndLogits
-    - forward
-
-[[autodoc]] modeling_utils.PoolerAnswerClass
-    - forward
-
-[[autodoc]] modeling_utils.SquadHeadOutput
-
-[[autodoc]] modeling_utils.SQuADHead
-    - forward
-
-[[autodoc]] modeling_utils.SequenceSummary
-    - forward
-
 ## PyTorch Helper Functions
 
 [[autodoc]] pytorch_utils.apply_chunking_to_forward

docs/source/en/kv_cache.md

@@ -18,7 +18,7 @@ rendered properly in your Markdown viewer.
 
 The key-value (KV) vectors are used to calculate attention scores. For autoregressive models, KV scores are calculated *every* time because the model predicts one token at a time. Each prediction depends on the previous tokens, which means the model performs the same computations each time.
 
-A KV *cache* stores these calculations so they can be reused without recomputing them. Efficient caching is crucial for optimizing model performance because it reduces computation time and improves response rates. Refer to the [Caching](./cache_explanation.md) doc for a more detailed explanation about how a cache works.
+A KV *cache* stores these calculations so they can be reused without recomputing them. Efficient caching is crucial for optimizing model performance because it reduces computation time and improves response rates. Refer to the [Caching](./cache_explanation) doc for a more detailed explanation about how a cache works.
 
 Transformers offers several [`Cache`] classes that implement different caching mechanisms. Some of these [`Cache`] classes are optimized to save memory while others are designed to maximize generation speed. Refer to the table below to compare cache types and use it to help you select the best cache for your use case.
 

docs/source/en/llm_optims.md

@@ -93,7 +93,7 @@ model.generation_config.max_new_tokens = 16
 
 past_key_values = StaticCache(
     config=model.config,
-    batch_size=1,
+    max_batch_size=1,
     # If you plan to reuse the cache, make sure the cache length is large enough for all cases
     max_cache_len=prompt_length+(model.generation_config.max_new_tokens*2),
     device=model.device,
@@ -159,7 +159,7 @@ from torch.nn.attention import SDPBackend, sdpa_kernel
 batch_size, seq_length = inputs["input_ids"].shape
 with torch.no_grad():
     past_key_values = StaticCache(
-        config=model.config, batch_size=2, max_cache_len=4096, device=torch_device, dtype=model.dtype
+        config=model.config, max_batch_size=2, max_cache_len=4096, device=torch_device, dtype=model.dtype
     )
     cache_position = torch.arange(seq_length, device=torch_device)
     generated_ids = torch.zeros(

docs/source/en/llm_tutorial_optimization.md

@@ -56,7 +56,7 @@ To give some examples of how much VRAM it roughly takes to load a model in bfloa
 
 As of writing this document, the largest GPU chip on the market is the A100 & H100 offering 80GB of VRAM. Most of the models listed before require more than 80GB just to be loaded and therefore necessarily require [tensor parallelism](https://huggingface.co/docs/transformers/perf_train_gpu_many#tensor-parallelism) and/or [pipeline parallelism](https://huggingface.co/docs/transformers/perf_train_gpu_many#naive-model-parallelism-vertical-and-pipeline-parallelism).
 
-🤗 Transformers now supports tensor parallelism for supported models having `base_tp_plan` in their respecitve config classes. Learn more about Tensor Parallelism [here](perf_train_gpu_many#tensor-parallelism). Furthermore, if you're interested in writing models in a tensor-parallelism-friendly way, feel free to have a look at [the text-generation-inference library](https://github.com/huggingface/text-generation-inference/tree/main/server/text_generation_server/models/custom_modeling).
+🤗 Transformers now supports tensor parallelism for supported models having `base_tp_plan` in their respective config classes. Learn more about Tensor Parallelism [here](perf_train_gpu_many#tensor-parallelism). Furthermore, if you're interested in writing models in a tensor-parallelism-friendly way, feel free to have a look at [the text-generation-inference library](https://github.com/huggingface/text-generation-inference/tree/main/server/text_generation_server/models/custom_modeling).
 
 Naive pipeline parallelism is supported out of the box. For this, simply load the model with `device="auto"` which will automatically place the different layers on the available GPUs as explained [here](https://huggingface.co/docs/accelerate/v0.22.0/en/concept_guides/big_model_inference).
 Note, however that while very effective, this naive pipeline parallelism does not tackle the issues of GPU idling. For this more advanced pipeline parallelism is required as explained [here](https://huggingface.co/docs/transformers/en/perf_train_gpu_many#naive-model-parallelism-vertical-and-pipeline-parallelism).
@@ -551,7 +551,7 @@ $$ \mathbf{\hat{q}}_i^T \mathbf{\hat{x}}_j = \mathbf{{q}}_i^T \mathbf{R}_{\theta
 
 \\( \mathbf{R}_{\theta, i - j} \\) thereby represents a rotational matrix. \\( \theta \\) is *not* learned during training, but instead set to a pre-defined value that depends on the maximum input sequence length during training.
 
-> By doing so, the propability score between \\( \mathbf{q}_i \\) and \\( \mathbf{q}_j \\) is only affected if \\( i \ne j \\) and solely depends on the relative distance \\( i - j \\) regardless of each vector's specific positions \\( i \\) and \\( j \\) .
+> By doing so, the probability score between \\( \mathbf{q}_i \\) and \\( \mathbf{q}_j \\) is only affected if \\( i \ne j \\) and solely depends on the relative distance \\( i - j \\) regardless of each vector's specific positions \\( i \\) and \\( j \\) .
 
 *RoPE* is used in multiple of today's most important LLMs, such as:
 

docs/source/en/main_classes/agent.md

@@ -1,167 +0,0 @@
-<!--Copyright 2023 The HuggingFace Team. All rights reserved.
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
-the License. You may obtain a copy of the License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
-an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.
-
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
-rendered properly in your Markdown viewer.
-
--->
-
-# Agents & Tools
-
-<Tip warning={true}>
-
-Transformers Agents is an experimental API which is subject to change at any time. Results returned by the agents
-can vary as the APIs or underlying models are prone to change.
-
-</Tip>
-
-To learn more about agents and tools make sure to read the [introductory guide](../transformers_agents). This page
-contains the API docs for the underlying classes.
-
-## Agents
-
-We provide two types of agents, based on the main [`Agent`] class:
-- [`CodeAgent`] acts in one shot, generating code to solve the task, then executes it at once.
-- [`ReactAgent`] acts step by step, each step consisting of one thought, then one tool call and execution. It has two classes:
-  - [`ReactJsonAgent`] writes its tool calls in JSON.
-  - [`ReactCodeAgent`] writes its tool calls in Python code.
-
-### Agent
-
-[[autodoc]] Agent
-
-### CodeAgent
-
-[[autodoc]] CodeAgent
-
-### React agents
-
-[[autodoc]] ReactAgent
-
-[[autodoc]] ReactJsonAgent
-
-[[autodoc]] ReactCodeAgent
-
-### ManagedAgent
-
-[[autodoc]] ManagedAgent
-
-## Tools
-
-### load_tool
-
-[[autodoc]] load_tool
-
-### tool
-
-[[autodoc]] tool
-
-### Tool
-
-[[autodoc]] Tool
-
-### Toolbox
-
-[[autodoc]] Toolbox
-
-### PipelineTool
-
-[[autodoc]] PipelineTool
-
-### launch_gradio_demo
-
-[[autodoc]] launch_gradio_demo
-
-### stream_to_gradio
-
-[[autodoc]] stream_to_gradio
-
-### ToolCollection
-
-[[autodoc]] ToolCollection
-
-## Engines
-
-You're free to create and use your own engines to be usable by the Agents framework.
-These engines have the following specification:
-1. Follow the [messages format](../chat_templating.md) for its input (`List[Dict[str, str]]`) and return a string.
-2. Stop generating outputs *before* the sequences passed in the argument `stop_sequences`
-
-### TransformersEngine
-
-For convenience, we have added a `TransformersEngine` that implements the points above, taking a pre-initialized `Pipeline` as input.
-
-```python
->>> from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, TransformersEngine
-
->>> model_name = "HuggingFaceTB/SmolLM-135M-Instruct"
->>> tokenizer = AutoTokenizer.from_pretrained(model_name)
->>> model = AutoModelForCausalLM.from_pretrained(model_name)
-
->>> pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
-
->>> engine = TransformersEngine(pipe)
->>> engine([{"role": "user", "content": "Ok!"}], stop_sequences=["great"])
-
-"What a "
-```
-
-[[autodoc]] TransformersEngine
-
-### HfApiEngine
-
-The `HfApiEngine` is an engine that wraps an [HF Inference API](https://huggingface.co/docs/api-inference/index) client for the execution of the LLM.
-
-```python
->>> from transformers import HfApiEngine
-
->>> messages = [
-...   {"role": "user", "content": "Hello, how are you?"},
-...   {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
-...   {"role": "user", "content": "No need to help, take it easy."},
-... ]
-
->>> HfApiEngine()(messages, stop_sequences=["conversation"])
-
-"That's very kind of you to say! It's always nice to have a relaxed "
-```
-
-[[autodoc]] HfApiEngine
-
-
-## Agent Types
-
-Agents can handle any type of object in-between tools; tools, being completely multimodal, can accept and return
-text, image, audio, video, among other types. In order to increase compatibility between tools, as well as to 
-correctly render these returns in ipython (jupyter, colab, ipython notebooks, ...), we implement wrapper classes
-around these types.
-
-The wrapped objects should continue behaving as initially; a text object should still behave as a string, an image
-object should still behave as a `PIL.Image`.
-
-These types have three specific purposes:
-
-- Calling `to_raw` on the type should return the underlying object
-- Calling `to_string` on the type should return the object as a string: that can be the string in case of an `AgentText`
-  but will be the path of the serialized version of the object in other instances
-- Displaying it in an ipython kernel should display the object correctly
-
-### AgentText
-
-[[autodoc]] transformers.agents.agent_types.AgentText
-
-### AgentImage
-
-[[autodoc]] transformers.agents.agent_types.AgentImage
-
-### AgentAudio
-
-[[autodoc]] transformers.agents.agent_types.AgentAudio

docs/source/en/main_classes/optimizer_schedules.md

@@ -22,9 +22,6 @@ The `.optimization` module provides:
 - several schedules in the form of schedule objects that inherit from `_LRSchedule`:
 - a gradient accumulation class to accumulate the gradients of multiple batches
 
-## AdamW (PyTorch)
-
-[[autodoc]] AdamW
 
 ## AdaFactor (PyTorch)
 

docs/source/en/main_classes/quantization.md

@@ -77,9 +77,9 @@ Learn how to quantize models in the [Quantization](../quantization) guide.
 
 [[autodoc]] TorchAoConfig
 
-## BitNetConfig
+## BitNetQuantConfig
 
-[[autodoc]] BitNetConfig
+[[autodoc]] BitNetQuantConfig
 
 ## SpQRConfig
 
@@ -88,3 +88,11 @@ Learn how to quantize models in the [Quantization](../quantization) guide.
 ## FineGrainedFP8Config
 
 [[autodoc]] FineGrainedFP8Config
+
+## QuarkConfig
+
+[[autodoc]] QuarkConfig
+
+## AutoRoundConfig
+
+[[autodoc]] AutoRoundConfig

docs/source/en/model_doc/audio-spectrogram-transformer.md

@@ -18,6 +18,7 @@ rendered properly in your Markdown viewer.
 
 <div class="flex flex-wrap space-x-1">
 <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
 <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
 </div>
 

docs/source/en/model_doc/bert.md

@@ -14,159 +14,85 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# BERT
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
-">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
+        ">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
+# BERT
 
-The BERT model was proposed in [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://arxiv.org/abs/1810.04805) by Jacob Devlin, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. It's a
-bidirectional transformer pretrained using a combination of masked language modeling objective and next sentence
-prediction on a large corpus comprising the Toronto Book Corpus and Wikipedia.
+[BERT](https://huggingface.co/papers/1810.04805) is a bidirectional transformer pretrained on unlabeled text to predict masked tokens in a sentence and to predict whether one sentence follows another. The main idea is that by randomly masking some tokens, the model can train on text to the left and right, giving it a more thorough understanding. BERT is also very versatile because its learned language representations can be adapted for other NLP tasks by fine-tuning an additional layer or head.
 
-The abstract from the paper is the following:
+You can find all the original BERT checkpoints under the [BERT](https://huggingface.co/collections/google/bert-release-64ff5e7a4be99045d1896dbc) collection.
 
-*We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations
-from Transformers. Unlike recent language representation models, BERT is designed to pre-train deep bidirectional
-representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result,
-the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models
-for a wide range of tasks, such as question answering and language inference, without substantial task-specific
-architecture modifications.*
+> [!TIP]
+> Click on the BERT models in the right sidebar for more examples of how to apply BERT to different language tasks.
 
-*BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural
-language processing tasks, including pushing the GLUE score to 80.5% (7.7% point absolute improvement), MultiNLI
-accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute
-improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).*
+The example below demonstrates how to predict the `[MASK]` token with [`Pipeline`], [`AutoModel`], and from the command line.
 
-This model was contributed by [thomwolf](https://huggingface.co/thomwolf). The original code can be found [here](https://github.com/google-research/bert).
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-## Usage tips
+```py
+import torch
+from transformers import pipeline
 
-- BERT is a model with absolute position embeddings so it's usually advised to pad the inputs on the right rather than
-  the left.
-- BERT was trained with the masked language modeling (MLM) and next sentence prediction (NSP) objectives. It is
-  efficient at predicting masked tokens and at NLU in general, but is not optimal for text generation.
-- Corrupts the inputs by using random masking, more precisely, during pretraining, a given percentage of tokens (usually 15%) is masked by:
-
-    * a special mask token with probability 0.8
-    * a random token different from the one masked with probability 0.1
-    * the same token with probability 0.1
-    
-- The model must predict the original sentence, but has a second objective: inputs are two sentences A and B (with a separation token in between). With probability 50%, the sentences are consecutive in the corpus, in the remaining 50% they are not related. The model has to predict if the sentences are consecutive or not.
-
-### Using Scaled Dot Product Attention (SDPA)
-
-PyTorch includes a native scaled dot-product attention (SDPA) operator as part of `torch.nn.functional`. This function 
-encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the 
-[official documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) 
-or the [GPU Inference](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention)
-page for more information.
-
-SDPA is used by default for `torch>=2.1.1` when an implementation is available, but you may also set 
-`attn_implementation="sdpa"` in `from_pretrained()` to explicitly request SDPA to be used.
-
-```
-from transformers import BertModel
-
-model = BertModel.from_pretrained("bert-base-uncased", torch_dtype=torch.float16, attn_implementation="sdpa")
-...
+pipeline = pipeline(
+    task="fill-mask",
+    model="google-bert/bert-base-uncased",
+    torch_dtype=torch.float16,
+    device=0
+)
+pipeline("Plants create [MASK] through a process known as photosynthesis.")
 ```
 
-For the best speedups, we recommend loading the model in half-precision (e.g. `torch.float16` or `torch.bfloat16`).
+</hfoption>
+<hfoption id="AutoModel">
 
-On a local benchmark (A100-80GB, CPUx12, RAM 96.6GB, PyTorch 2.2.0, OS Ubuntu 22.04) with `float16`, we saw the 
-following speedups during training and inference.
+```py
+import torch
+from transformers import AutoModelForMaskedLM, AutoTokenizer
 
-#### Training
+tokenizer = AutoTokenizer.from_pretrained(
+    "google-bert/bert-base-uncased",
+)
+model = AutoModelForMaskedLM.from_pretrained(
+    "google-bert/bert-base-uncased",
+    torch_dtype=torch.float16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
+inputs = tokenizer("Plants create [MASK] through a process known as photosynthesis.", return_tensors="pt").to("cuda")
 
-|batch_size|seq_len|Time per batch (eager - s)|Time per batch (sdpa - s)|Speedup (%)|Eager peak mem (MB)|sdpa peak mem (MB)|Mem saving (%)|
-|----------|-------|--------------------------|-------------------------|-----------|-------------------|------------------|--------------|
-|4         |256    |0.023                     |0.017                    |35.472     |939.213            |764.834           |22.800        |
-|4         |512    |0.023                     |0.018                    |23.687     |1970.447           |1227.162          |60.569        |
-|8         |256    |0.023                     |0.018                    |23.491     |1594.295           |1226.114          |30.028        |
-|8         |512    |0.035                     |0.025                    |43.058     |3629.401           |2134.262          |70.054        |
-|16        |256    |0.030                     |0.024                    |25.583     |2874.426           |2134.262          |34.680        |
-|16        |512    |0.064                     |0.044                    |46.223     |6964.659           |3961.013          |75.830        |
+with torch.no_grad():
+    outputs = model(**inputs)
+    predictions = outputs.logits
 
-#### Inference
+masked_index = torch.where(inputs['input_ids'] == tokenizer.mask_token_id)[1]
+predicted_token_id = predictions[0, masked_index].argmax(dim=-1)
+predicted_token = tokenizer.decode(predicted_token_id)
 
-|batch_size|seq_len|Per token latency eager (ms)|Per token latency SDPA (ms)|Speedup (%)|Mem eager (MB)|Mem BT (MB)|Mem saved (%)|
-|----------|-------|----------------------------|---------------------------|-----------|--------------|-----------|-------------|
-|1         |128    |5.736                       |4.987                      |15.022     |282.661       |282.924    |-0.093       |
-|1         |256    |5.689                       |4.945                      |15.055     |298.686       |298.948    |-0.088       |
-|2         |128    |6.154                       |4.982                      |23.521     |314.523       |314.785    |-0.083       |
-|2         |256    |6.201                       |4.949                      |25.303     |347.546       |347.033    |0.148        |
-|4         |128    |6.049                       |4.987                      |21.305     |378.895       |379.301    |-0.107       |
-|4         |256    |6.285                       |5.364                      |17.166     |443.209       |444.382    |-0.264       |
+print(f"The predicted token is: {predicted_token}")
+```
 
+</hfoption>
+<hfoption id="transformers-cli">
 
+```bash
+echo -e "Plants create [MASK] through a process known as photosynthesis." | transformers-cli run --task fill-mask --model google-bert/bert-base-uncased --device 0
+```
 
-## Resources
+</hfoption>
+</hfoptions>
 
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with BERT. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
+## Notes
 
-<PipelineTag pipeline="text-classification"/>
-
-- A blog post on [BERT Text Classification in a different language](https://www.philschmid.de/bert-text-classification-in-a-different-language).
-- A notebook for [Finetuning BERT (and friends) for multi-label text classification](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/BERT/Fine_tuning_BERT_(and_friends)_for_multi_label_text_classification.ipynb).
-- A notebook on how to [Finetune BERT for multi-label classification using PyTorch](https://colab.research.google.com/github/abhimishra91/transformers-tutorials/blob/master/transformers_multi_label_classification.ipynb). 🌎
-- A notebook on how to [warm-start an EncoderDecoder model with BERT for summarization](https://colab.research.google.com/github/patrickvonplaten/notebooks/blob/master/BERT2BERT_for_CNN_Dailymail.ipynb).
-- [`BertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification.ipynb).
-- [`TFBertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification-tf.ipynb).
-- [`FlaxBertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification_flax.ipynb).
-- [Text classification task guide](../tasks/sequence_classification)
-
-<PipelineTag pipeline="token-classification"/>
-
-- A blog post on how to use [Hugging Face Transformers with Keras: Fine-tune a non-English BERT for Named Entity Recognition](https://www.philschmid.de/huggingface-transformers-keras-tf).
-- A notebook for [Finetuning BERT for named-entity recognition](https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/BERT/Custom_Named_Entity_Recognition_with_BERT_only_first_wordpiece.ipynb) using only the first wordpiece of each word in the word label during tokenization. To propagate the label of the word to all wordpieces, see this [version](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/BERT/Custom_Named_Entity_Recognition_with_BERT.ipynb) of the notebook instead.
-- [`BertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification.ipynb).
-- [`TFBertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification-tf.ipynb).
-- [`FlaxBertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/token-classification).
-- [Token classification](https://huggingface.co/course/chapter7/2?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Token classification task guide](../tasks/token_classification)
-
-<PipelineTag pipeline="fill-mask"/>
-
-- [`BertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#robertabertdistilbert-and-masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
-- [`TFBertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling#run_mlmpy) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling-tf.ipynb).
-- [`FlaxBertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/language-modeling#masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/masked_language_modeling_flax.ipynb).
-- [Masked language modeling](https://huggingface.co/course/chapter7/3?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Masked language modeling task guide](../tasks/masked_language_modeling)
-
-<PipelineTag pipeline="question-answering"/>
-
-- [`BertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering.ipynb).
-- [`TFBertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering-tf.ipynb).
-- [`FlaxBertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/question-answering).
-- [Question answering](https://huggingface.co/course/chapter7/7?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Question answering task guide](../tasks/question_answering)
-
-**Multiple choice**
-- [`BertForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice.ipynb).
-- [`TFBertForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice-tf.ipynb).
-- [Multiple choice task guide](../tasks/multiple_choice)
-
-⚡️ **Inference**
-- A blog post on how to [Accelerate BERT inference with Hugging Face Transformers and AWS Inferentia](https://huggingface.co/blog/bert-inferentia-sagemaker).
-- A blog post on how to [Accelerate BERT inference with DeepSpeed-Inference on GPUs](https://www.philschmid.de/bert-deepspeed-inference).
-
-⚙️ **Pretraining**
-- A blog post on [Pre-Training BERT with Hugging Face Transformers and Habana Gaudi](https://www.philschmid.de/pre-training-bert-habana).
-
-🚀 **Deploy**
-- A blog post on how to [Convert Transformers to ONNX with Hugging Face Optimum](https://www.philschmid.de/convert-transformers-to-onnx).
-- A blog post on how to [Setup Deep Learning environment for Hugging Face Transformers with Habana Gaudi on AWS](https://www.philschmid.de/getting-started-habana-gaudi#conclusion).
-- A blog post on [Autoscaling BERT with Hugging Face Transformers, Amazon SageMaker and Terraform module](https://www.philschmid.de/terraform-huggingface-amazon-sagemaker-advanced).
-- A blog post on [Serverless BERT with HuggingFace, AWS Lambda, and Docker](https://www.philschmid.de/serverless-bert-with-huggingface-aws-lambda-docker).
-- A blog post on [Hugging Face Transformers BERT fine-tuning using Amazon SageMaker and Training Compiler](https://www.philschmid.de/huggingface-amazon-sagemaker-training-compiler).
-- A blog post on [Task-specific knowledge distillation for BERT using Transformers & Amazon SageMaker](https://www.philschmid.de/knowledge-distillation-bert-transformers).
+- Inputs should be padded on the right because BERT uses absolute position embeddings.
 
 ## BertConfig
 
@@ -181,35 +107,10 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
     - create_token_type_ids_from_sequences
     - save_vocabulary
 
-<frameworkcontent>
-<pt>
-
 ## BertTokenizerFast
 
 [[autodoc]] BertTokenizerFast
 
-</pt>
-<tf>
-
-## TFBertTokenizer
-
-[[autodoc]] TFBertTokenizer
-
-</tf>
-</frameworkcontent>
-
-## Bert specific outputs
-
-[[autodoc]] models.bert.modeling_bert.BertForPreTrainingOutput
-
-[[autodoc]] models.bert.modeling_tf_bert.TFBertForPreTrainingOutput
-
-[[autodoc]] models.bert.modeling_flax_bert.FlaxBertForPreTrainingOutput
-
-
-<frameworkcontent>
-<pt>
-
 ## BertModel
 
 [[autodoc]] BertModel
@@ -255,8 +156,9 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 [[autodoc]] BertForQuestionAnswering
     - forward
 
-</pt>
-<tf>
+## TFBertTokenizer
+
+[[autodoc]] TFBertTokenizer
 
 ## TFBertModel
 
@@ -303,9 +205,6 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 [[autodoc]] TFBertForQuestionAnswering
     - call
 
-</tf>
-<jax>
-
 ## FlaxBertModel
 
 [[autodoc]] FlaxBertModel
@@ -351,7 +250,10 @@ A list of official Hugging Face and community (indicated by 🌎) resources to h
 [[autodoc]] FlaxBertForQuestionAnswering
     - __call__
 
-</jax>
-</frameworkcontent>
+## Bert specific outputs
 
+[[autodoc]] models.bert.modeling_bert.BertForPreTrainingOutput
 
+[[autodoc]] models.bert.modeling_tf_bert.TFBertForPreTrainingOutput
+
+[[autodoc]] models.bert.modeling_flax_bert.FlaxBertForPreTrainingOutput

docs/source/en/model_doc/bit.md

@@ -58,6 +58,11 @@ If you're interested in submitting a resource to be included here, please feel f
 [[autodoc]] BitImageProcessor
     - preprocess
 
+## BitImageProcessorFast
+
+[[autodoc]] BitImageProcessorFast
+    - preprocess
+
 ## BitModel
 
 [[autodoc]] BitModel

docs/source/en/model_doc/bitnet.md

@@ -0,0 +1,121 @@
+<!--Copyright 2025 The BitNet Team and The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# BitNet
+
+## Overview
+
+Trained on a corpus of 4 trillion tokens, this model demonstrates that native 1-bit LLMs can achieve performance comparable to leading open-weight, full-precision models of similar size, while offering substantial advantages in computational efficiency (memory, energy, latency).
+
+➡️ **Technical Report:** [BitNet b1.58 2B4T Technical Report](https://arxiv.org/abs/2504.12285)
+
+➡️ **Official Inference Code:** [microsoft/BitNet (bitnet.cpp)](https://github.com/microsoft/BitNet)
+
+## Model Variants
+
+Several versions of the model weights are available on Hugging Face:
+
+* [**`microsoft/bitnet-b1.58-2B-4T`**](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T): Contains the packed 1.58-bit weights optimized for efficient inference. **Use this for deployment.**
+
+* [**`microsoft/bitnet-b1.58-2B-4T-bf16`**](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-bf16): Contains the master weights in BF16 format. **Use this only for training or fine-tuning purposes.**
+
+* [**`microsoft/bitnet-b1.58-2B-4T-gguf`**](https://huggingface.co/microsoft/bitnet-b1.58-2B-4T-gguf): Contains the model weights in GGUF format, compatible with the `bitnet.cpp` library for CPU inference.
+
+
+### Model Details
+
+
+* **Architecture:** Transformer-based, modified with `BitLinear` layers (BitNet framework).
+    * Uses Rotary Position Embeddings (RoPE).
+    * Uses squared ReLU (ReLU²) activation in FFN layers.
+    * Employs [`subln`](https://proceedings.mlr.press/v202/wang23u.html) normalization.
+    * No bias terms in linear or normalization layers.
+* **Quantization:** Native 1.58-bit weights and 8-bit activations (W1.58A8).
+    * Weights are quantized to ternary values {-1, 0, +1} using absmean quantization during the forward pass.
+    * Activations are quantized to 8-bit integers using absmax quantization (per-token).
+    * **Crucially, the model was *trained from scratch* with this quantization scheme, not post-training quantized.**
+* **Parameters:** ~2 Billion
+* **Training Tokens:** 4 Trillion
+*   **Context Length:** Maximum sequence length of **4096 tokens**.
+    *   *Recommendation:* For optimal performance on tasks requiring very long contexts (beyond the pre-training length or for specialized long-reasoning tasks), we recommend performing intermediate long-sequence adaptation/training before the final fine-tuning stage.
+* **Training Stages:**
+    1.  **Pre-training:** Large-scale training on public text/code and synthetic math data using a two-stage learning rate and weight decay schedule.
+    2.  **Supervised Fine-tuning (SFT):** Fine-tuned on instruction-following and conversational datasets using sum loss aggregation and specific hyperparameter tuning.
+    3.  **Direct Preference Optimization (DPO):** Aligned with human preferences using preference pairs.
+* **Tokenizer:** LLaMA 3 Tokenizer (vocab size: 128,256).
+
+
+## Usage tips
+
+
+**VERY IMPORTANT NOTE ON EFFICIENCY**
+
+> Please do NOT expect performance efficiency gains (in terms of speed, latency, or energy consumption) when using this model with the standard transformers library.
+>
+> The current execution paths within transformers do not contain the specialized, highly optimized computational kernels required to leverage the advantages of the BitNet architecture. Running the model via transformers will likely result in inference speeds and energy usage comparable to, or potentially worse than, standard full-precision models within this framework on both CPU and GPU.
+>
+> While you might observe reduced memory usage due to the quantized weights, the primary computational efficiency benefits are not accessible through this standard transformers usage path.
+>
+> For achieving the efficiency benefits demonstrated in the technical paper, you MUST use the dedicated C++ implementation: [bitnet.cpp](https://github.com/microsoft/BitNet).
+
+### Requirements
+
+```bash
+pip install transformers
+```
+
+### Example
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+model_id = "microsoft/bitnet-b1.58-2B-4T"
+
+# Load tokenizer and model
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16
+)
+
+# Apply the chat template
+messages = [
+    {"role": "system", "content": "You are a helpful AI assistant."},
+    {"role": "user", "content": "How are you?"},
+]
+chat_input = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
+
+# Generate response
+chat_outputs = model.generate(chat_input, max_new_tokens=50)
+response = tokenizer.decode(chat_outputs[0][chat_input.shape[-1]:], skip_special_tokens=True) # Decode only the response part
+print("\nAssistant Response:", response)
+```
+
+
+## BitNetConfig
+
+[[autodoc]] BitNetConfig
+
+## BitNetModel
+
+[[autodoc]] BitNetModel
+    - forward
+
+## BitNetForCausalLM
+
+[[autodoc]] BitNetForCausalLM
+    - forward

docs/source/en/model_doc/blip.md

@@ -88,6 +88,11 @@ The original code can be found [here](https://github.com/salesforce/BLIP).
 [[autodoc]] BlipTextModel
     - forward
 
+## BlipTextLMHeadModel
+
+[[autodoc]] BlipTextLMHeadModel
+- forward
+
 ## BlipVisionModel
 
 [[autodoc]] BlipVisionModel
@@ -123,6 +128,11 @@ The original code can be found [here](https://github.com/salesforce/BLIP).
 [[autodoc]] TFBlipTextModel
     - call
 
+## TFBlipTextLMHeadModel
+
+[[autodoc]] TFBlipTextLMHeadModel
+- forward
+
 ## TFBlipVisionModel
 
 [[autodoc]] TFBlipVisionModel

docs/source/en/model_doc/bridgetower.md

@@ -147,6 +147,11 @@ Tips:
 [[autodoc]] BridgeTowerImageProcessor
     - preprocess
 
+## BridgeTowerImageProcessorFast
+
+[[autodoc]] BridgeTowerImageProcessorFast
+    - preprocess
+
 ## BridgeTowerProcessor
 
 [[autodoc]] BridgeTowerProcessor

docs/source/en/model_doc/chinese_clip.md

@@ -90,6 +90,11 @@ Currently, following scales of pretrained Chinese-CLIP models are available on
 [[autodoc]] ChineseCLIPImageProcessor
     - preprocess
 
+## ChineseCLIPImageProcessorFast
+
+[[autodoc]] ChineseCLIPImageProcessorFast
+    - preprocess
+
 ## ChineseCLIPFeatureExtractor
 
 [[autodoc]] ChineseCLIPFeatureExtractor

docs/source/en/model_doc/clip.md

@@ -14,221 +14,77 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# CLIP
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
-">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
+        ">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
+# CLIP
 
-The CLIP model was proposed in [Learning Transferable Visual Models From Natural Language Supervision](https://arxiv.org/abs/2103.00020) by Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh,
-Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, Gretchen Krueger, Ilya Sutskever. CLIP
-(Contrastive Language-Image Pre-Training) is a neural network trained on a variety of (image, text) pairs. It can be
-instructed in natural language to predict the most relevant text snippet, given an image, without directly optimizing
-for the task, similarly to the zero-shot capabilities of GPT-2 and 3.
+[CLIP](https://huggingface.co/papers/2103.00020) is a is a multimodal vision and language model motivated by overcoming the fixed number of object categories when training a computer vision model. CLIP learns about images directly from raw text by jointly training on 400M (image, text) pairs. Pretraining on this scale enables zero-shot transfer to downstream tasks. CLIP uses an image encoder and text encoder to get visual features and text features. Both features are projected to a latent space with the same number of dimensions and their dot product gives a similarity score.
 
-The abstract from the paper is the following:
+You can find all the original CLIP checkpoints under the [OpenAI](https://huggingface.co/openai?search_models=clip) organization.
 
-*State-of-the-art computer vision systems are trained to predict a fixed set of predetermined object categories. This
-restricted form of supervision limits their generality and usability since additional labeled data is needed to specify
-any other visual concept. Learning directly from raw text about images is a promising alternative which leverages a
-much broader source of supervision. We demonstrate that the simple pre-training task of predicting which caption goes
-with which image is an efficient and scalable way to learn SOTA image representations from scratch on a dataset of 400
-million (image, text) pairs collected from the internet. After pre-training, natural language is used to reference
-learned visual concepts (or describe new ones) enabling zero-shot transfer of the model to downstream tasks. We study
-the performance of this approach by benchmarking on over 30 different existing computer vision datasets, spanning tasks
-such as OCR, action recognition in videos, geo-localization, and many types of fine-grained object classification. The
-model transfers non-trivially to most tasks and is often competitive with a fully supervised baseline without the need
-for any dataset specific training. For instance, we match the accuracy of the original ResNet-50 on ImageNet zero-shot
-without needing to use any of the 1.28 million training examples it was trained on. We release our code and pre-trained
-model weights at this https URL.*
+> [!TIP]
+> Click on the CLIP models in the right sidebar for more examples of how to apply CLIP to different image and language tasks.
 
-This model was contributed by [valhalla](https://huggingface.co/valhalla). The original code can be found [here](https://github.com/openai/CLIP).
+The example below demonstrates how to calculate similarity scores between multiple text descriptions and an image with [`Pipeline`] or the [`AutoModel`] class.
 
-## Usage tips and example
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-CLIP is a multi-modal vision and language model. It can be used for image-text similarity and for zero-shot image
-classification. CLIP uses a ViT like transformer to get visual features and a causal language model to get the text
-features. Both the text and visual features are then projected to a latent space with identical dimension. The dot
-product between the projected image and text features is then used as a similar score.
+```py
+import torch
+from transformers import pipeline
 
-To feed images to the Transformer encoder, each image is split into a sequence of fixed-size non-overlapping patches,
-which are then linearly embedded. A [CLS] token is added to serve as representation of an entire image. The authors
-also add absolute position embeddings, and feed the resulting sequence of vectors to a standard Transformer encoder.
-The [`CLIPImageProcessor`] can be used to resize (or rescale) and normalize images for the model.
-
-The [`CLIPTokenizer`] is used to encode the text. The [`CLIPProcessor`] wraps
-[`CLIPImageProcessor`] and [`CLIPTokenizer`] into a single instance to both
-encode the text and prepare the images. The following example shows how to get the image-text similarity scores using
-[`CLIPProcessor`] and [`CLIPModel`].
-
-
-```python
->>> from PIL import Image
->>> import requests
-
->>> from transformers import CLIPProcessor, CLIPModel
-
->>> model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
->>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
-
->>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
->>> image = Image.open(requests.get(url, stream=True).raw)
-
->>> inputs = processor(text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True)
-
->>> outputs = model(**inputs)
->>> logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
->>> probs = logits_per_image.softmax(dim=1)  # we can take the softmax to get the label probabilities
+clip = pipeline(
+   task="zero-shot-image-classification",
+   model="openai/clip-vit-base-patch32",
+   torch_dtype=torch.bfloat16,
+   device=0
+)
+labels = ["a photo of a cat", "a photo of a dog", "a photo of a car"]
+clip("http://images.cocodataset.org/val2017/000000039769.jpg", candidate_labels=labels)
 ```
 
+</hfoption>
+<hfoption id="AutoModel">
 
-### Combining CLIP and Flash Attention 2
+```py
+import requests
+import torch
+from PIL import Image
+from transformers import AutoProcessor, AutoModel
 
-First, make sure to install the latest version of Flash Attention 2.
+model = AutoModel.from_pretrained("openai/clip-vit-base-patch32", torch_dtype=torch.bfloat16, attn_implementation="sdpa")
+processor = AutoProcessor.from_pretrained("openai/clip-vit-base-patch32")
 
-```bash
-pip install -U flash-attn --no-build-isolation
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+labels = ["a photo of a cat", "a photo of a dog", "a photo of a car"]
+
+inputs = processor(text=labels, images=image, return_tensors="pt", padding=True)
+
+outputs = model(**inputs)
+logits_per_image = outputs.logits_per_image
+probs = logits_per_image.softmax(dim=1)
+most_likely_idx = probs.argmax(dim=1).item()
+most_likely_label = labels[most_likely_idx]
+print(f"Most likely label: {most_likely_label} with probability: {probs[0][most_likely_idx].item():.3f}")
 ```
 
-Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of flash-attn repository. Make also sure to load your model in half-precision (e.g. `torch.float16`)
+</hfoption>
+</hfoptions>
 
-<Tip warning={true}>
+## Notes
 
-For small batch sizes, you might notice a slowdown in your model when using flash attention. Refer to the section [Expected speedups with Flash Attention and SDPA](#Expected-speedups-with-Flash-Attention-and-SDPA) below and select an appropriate attention implementation.
-
-</Tip>
-
-To load and run a model using Flash Attention 2, refer to the snippet below:
-
-```python
->>> import torch
->>> import requests
->>> from PIL import Image
-
->>> from transformers import CLIPProcessor, CLIPModel
-
->>> device = "cuda"
->>> torch_dtype = torch.float16
-
->>> model = CLIPModel.from_pretrained(
-...     "openai/clip-vit-base-patch32",
-...     attn_implementation="flash_attention_2",
-...     device_map=device,
-...     torch_dtype=torch_dtype,
-... )
->>> processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
-
->>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
->>> image = Image.open(requests.get(url, stream=True).raw)
-
->>> inputs = processor(text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True)
->>> inputs.to(device)
-
->>> with torch.no_grad():
-...     with torch.autocast(device):
-...         outputs = model(**inputs)
-
->>> logits_per_image = outputs.logits_per_image  # this is the image-text similarity score
->>> probs = logits_per_image.softmax(dim=1)  # we can take the softmax to get the label probabilities
->>> print(probs)
-tensor([[0.9946, 0.0052]], device='cuda:0', dtype=torch.float16)
-```
-
-
-### Using Scaled Dot Product Attention (SDPA)
-
-PyTorch includes a native scaled dot-product attention (SDPA) operator as part of `torch.nn.functional`. This function 
-encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the 
-[official documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) 
-or the [GPU Inference](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention)
-page for more information.
-
-SDPA is used by default for `torch>=2.1.1` when an implementation is available, but you may also set 
-`attn_implementation="sdpa"` in `from_pretrained()` to explicitly request SDPA to be used.
-
-```python
-from transformers import CLIPModel
-
-model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32", torch_dtype=torch.float16, attn_implementation="sdpa")
-```
-
-For the best speedups, we recommend loading the model in half-precision (e.g. `torch.float16` or `torch.bfloat16`).
-
-### Expected speedups with Flash Attention and SDPA
-
-On a local benchmark (NVIDIA A10G, PyTorch 2.3.1+cu121) with `float16`, we saw the following speedups during inference for `"openai/clip-vit-large-patch14"` checkpoint ([code](https://gist.github.com/qubvel/ac691a54e54f9fae8144275f866a7ff8)):
-
-#### CLIPTextModel
-
-|   Num text labels |   Eager (s/iter) |   FA2 (s/iter) |   FA2 speedup |   SDPA (s/iter) |   SDPA speedup |
-|------------------:|-----------------:|---------------:|--------------:|----------------:|---------------:|
-|                 4 |            0.009 |          0.012 |         0.737 |           0.007 |          1.269 |
-|                16 |            0.009 |          0.014 |         0.659 |           0.008 |          1.187 |
-|                32 |            0.018 |          0.021 |         0.862 |           0.016 |          1.142 |
-|                64 |            0.034 |          0.034 |         1.001 |           0.03  |          1.163 |
-|               128 |            0.063 |          0.058 |         1.09  |           0.054 |          1.174 |
-
-![clip_text_model_viz_3](https://github.com/user-attachments/assets/e9826b43-4e66-4f4c-952b-af4d90bd38eb)
-
-#### CLIPVisionModel
-
-|   Image batch size |   Eager (s/iter) |   FA2 (s/iter) |   FA2 speedup |   SDPA (s/iter) |   SDPA speedup |
-|-------------------:|-----------------:|---------------:|--------------:|----------------:|---------------:|
-|                  1 |            0.016 |          0.013 |         1.247 |           0.012 |          1.318 |
-|                  4 |            0.025 |          0.021 |         1.198 |           0.021 |          1.202 |
-|                 16 |            0.093 |          0.075 |         1.234 |           0.075 |          1.24  |
-|                 32 |            0.181 |          0.147 |         1.237 |           0.146 |          1.241 |
-
-![clip_image_model_viz_3](https://github.com/user-attachments/assets/50a36206-e3b9-4adc-ac8e-926b8b071d63)
-
-#### CLIPModel
-
-|   Image batch size |   Num text labels |   Eager (s/iter) |   FA2 (s/iter) |   FA2 speedup |   SDPA (s/iter) |   SDPA speedup |
-|-------------------:|------------------:|-----------------:|---------------:|--------------:|----------------:|---------------:|
-|                  1 |                 4 |            0.025 |          0.026 |         0.954 |           0.02  |          1.217 |
-|                  1 |                16 |            0.026 |          0.028 |         0.918 |           0.02  |          1.287 |
-|                  1 |                64 |            0.042 |          0.046 |         0.906 |           0.036 |          1.167 |
-|                  4 |                 4 |            0.028 |          0.033 |         0.849 |           0.024 |          1.189 |
-|                  4 |                16 |            0.034 |          0.035 |         0.955 |           0.029 |          1.169 |
-|                  4 |                64 |            0.059 |          0.055 |         1.072 |           0.05  |          1.179 |
-|                 16 |                 4 |            0.096 |          0.088 |         1.091 |           0.078 |          1.234 |
-|                 16 |                16 |            0.102 |          0.09  |         1.129 |           0.083 |          1.224 |
-|                 16 |                64 |            0.127 |          0.11  |         1.157 |           0.105 |          1.218 |
-|                 32 |                 4 |            0.185 |          0.159 |         1.157 |           0.149 |          1.238 |
-|                 32 |                16 |            0.19  |          0.162 |         1.177 |           0.154 |          1.233 |
-|                 32 |                64 |            0.216 |          0.181 |         1.19  |           0.176 |          1.228 |
-
-## Resources
-
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with CLIP.
-
-- [Fine tuning CLIP with Remote Sensing (Satellite) images and captions](https://huggingface.co/blog/fine-tune-clip-rsicd), a blog post about how to fine-tune CLIP with [RSICD dataset](https://github.com/201528014227051/RSICD_optimal) and comparison of performance changes due to data augmentation.
-- This [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/contrastive-image-text) shows how to train a CLIP-like vision-text dual encoder model using a pre-trained vision and text encoder using [COCO dataset](https://cocodataset.org/#home).
-
-<PipelineTag pipeline="image-to-text"/>
-
-- A [notebook](https://colab.research.google.com/drive/1tuoAC5F4sC7qid56Z0ap-stR3rwdk0ZV?usp=sharing) on how to use a pretrained CLIP for inference with beam search for image captioning. 🌎
-
-**Image retrieval**
-
-- A [notebook](https://colab.research.google.com/drive/1bLVwVKpAndpEDHqjzxVPr_9nGrSbuOQd?usp=sharing) on image retrieval using pretrained CLIP and computing MRR(Mean Reciprocal Rank) score. 🌎
-- A [notebook](https://colab.research.google.com/github/deep-diver/image_search_with_natural_language/blob/main/notebooks/Image_Search_CLIP.ipynb) on image retrieval and showing the similarity score. 🌎
-- A [notebook](https://colab.research.google.com/drive/1xO-wC_m_GNzgjIBQ4a4znvQkvDoZJvH4?usp=sharing) on how to map images and texts to the same vector space using Multilingual CLIP. 🌎 
-- A [notebook](https://colab.research.google.com/github/vivien000/clip-demo/blob/master/clip.ipynb#scrollTo=uzdFhRGqiWkR) on how to run CLIP on semantic image search using [Unsplash](https://unsplash.com) and [TMDB](https://www.themoviedb.org/) datasets. 🌎
-
-**Explainability**
-
-- A [notebook](https://colab.research.google.com/github/hila-chefer/Transformer-MM-Explainability/blob/main/CLIP_explainability.ipynb) on how to visualize similarity between input token and image segment. 🌎
-
-If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we will review it.
-The resource should ideally demonstrate something new instead of duplicating an existing resource.
+- Use [`CLIPImageProcessor`] to resize (or rescale) and normalizes images for the model.
 
 ## CLIPConfig
 

docs/source/en/model_doc/code_llama.md

@@ -14,108 +14,154 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# CodeLlama
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAC0AAAAtCAMAAAANxBKoAAAC7lBMVEUAAADg5vYHPVgAoJH+/v76+v39/f9JbLP///9+AIgAnY3///+mcqzt8fXy9fgkXa3Ax9709fr+///9/f8qXq49qp5AaLGMwrv8/P0eW60VWawxYq8yqJzG2dytt9Wyu9elzci519Lf3O3S2efY3OrY0+Xp7PT///////+dqNCexMc6Z7AGpJeGvbenstPZ5ejQ1OfJzOLa7ejh4+/r8fT29vpccbklWK8PVa0AS6ghW63O498vYa+lsdKz1NDRt9Kw1c672tbD3tnAxt7R6OHp5vDe7OrDyuDn6vLl6/EAQKak0MgATakkppo3ZK/Bz9y8w9yzu9jey97axdvHzeG21NHH4trTwthKZrVGZLSUSpuPQJiGAI+GAI8SWKydycLL4d7f2OTi1+S9xNzL0ePT6OLGzeEAo5U0qJw/aLEAo5JFa7JBabEAp5Y4qZ2QxLyKmsm3kL2xoMOehrRNb7RIbbOZgrGre68AUqwAqZqNN5aKJ5N/lMq+qsd8kMa4pcWzh7muhLMEV69juq2kbKqgUaOTR5uMMZWLLZSGAI5VAIdEAH+ovNDHuNCnxcy3qcaYx8K8msGplrx+wLahjbYdXrV6vbMvYK9DrZ8QrZ8tqJuFms+Sos6sw8ecy8RffsNVeMCvmb43aLltv7Q4Y7EZWK4QWa1gt6meZKUdr6GOAZVeA4xPAISyveLUwtivxtKTpNJ2jcqfvcltiMiwwcfAoMVxhL+Kx7xjdrqTe60tsaNQs6KaRKACrJ6UTZwkqpqTL5pkHY4AloSgsd2ptNXPvNOOncuxxsqFl8lmg8apt8FJcr9EbryGxLqlkrkrY7dRa7ZGZLQ5t6iXUZ6PPpgVpZeJCJFKAIGareTa0+KJod3H0deY2M+esM25usmYu8d2zsJOdcBVvrCLbqcAOaaHaKQAMaScWqKBXqCXMJ2RHpiLF5NmJZAdAHN2kta11dKu1M+DkcZLdb+Mcql3TppyRJdzQ5ZtNZNlIY+DF4+voCOQAAAAZ3RSTlMABAT+MEEJ/RH+/TP+Zlv+pUo6Ifz8+fco/fz6+evr39S9nJmOilQaF/7+/f38+smmoYp6b1T+/v7++vj189zU0tDJxsGzsrKSfv34+Pf27dDOysG9t6+n/vv6+vr59uzr1tG+tZ6Qg9Ym3QAABR5JREFUSMeNlVVUG1EQhpcuxEspXqS0SKEtxQp1d3d332STTRpIQhIISQgJhODu7lAoDoUCpe7u7u7+1puGpqnCPOyZvffbOXPm/PsP9JfQgyCC+tmTABTOcbxDz/heENS7/1F+9nhvkHePG0wNDLbGWwdXL+rbLWvpmZHXD8+gMfBjTh+aSe6Gnn7lwQIOTR0c8wfX3PWgv7avbdKwf/ZoBp1Gp/PvuvXW3vw5ib7emnTW4OR+3D4jB9vjNJ/7gNvfWWeH/TO/JyYrsiKCRjVEZA3UB+96kON+DxOQ/NLE8PE5iUYgIXjFnCOlxEQMaSGVxjg4gxOnEycGz8bptuNjVx08LscIgrzH3umcn+KKtiBIyvzOO2O99aAdR8cF19oZalnCtvREUw79tCd5sow1g1UKM6kXqUx4T8wsi3sTjJ3yzDmmhenLXLpo8u45eG5y4Vvbk6kkC4LLtJMowkSQxmk4ggVJEG+7c6QpHT8vvW9X7/o7+3ELmiJi2mEzZJiz8cT6TBlanBk70cB5GGIGC1gRDdZ00yADLW1FL6gqhtvNXNG5S9gdSrk4M1qu7JAsmYshzDS4peoMrU/gT7qQdqYGZaYhxZmVbGJAm/CS/HloWyhRUlknQ9KYcExTwS80d3VNOxUZJpITYyspl0LbhArhpZCD9cRWEQuhYkNGMHToQ/2Cs6swJlb39CsllxdXX6IUKh/H5jbnSsPKjgmoaFQ1f8wRLR0UnGE/RcDEjj2jXG1WVTwUs8+zxfcrVO+vSsuOpVKxCfYZiQ0/aPKuxQbQ8lIz+DClxC8u+snlcJ7Yr1z1JPqUH0V+GDXbOwAib931Y4Imaq0NTIXPXY+N5L18GJ37SVWu+hwXff8l72Ds9XuwYIBaXPq6Shm4l+Vl/5QiOlV+uTk6YR9PxKsI9xNJny31ygK1e+nIRC1N97EGkFPI+jCpiHe5PCEy7oWqWSwRrpOvhFzcbTWMbm3ZJAOn1rUKpYIt/lDhW/5RHHteeWFN60qo98YJuoq1nK3uW5AabyspC1BcIEpOhft+SZAShYoLSvnmSfnYADUERP5jJn2h5XtsgCRuhYQqAvwTwn33+YWEKUI72HX5AtfSAZDe8F2DtPPm77afhl0EkthzuCQU0BWApgQIH9+KB0JhopMM7bJrdTRoleM2JAVNMyPF+wdoaz+XJpGoVAQ7WXUkcV7gT3oUZyi/ISIJAVKhgNp+4b4veCFhYVJw4locdSjZCp9cPUhLF9EZ3KKzURepMEtCDPP3VcWFx4UIiZIklIpFNfHpdEafIF2aRmOcrUmjohbT2WUllbmRvgfbythbQO3222fpDJoufaQPncYYuqoGtUEsCJZL6/3PR5b4syeSjZMQG/T2maGANlXT2v8S4AULWaUkCxfLyW8iW4kdka+nEMjxpL2NCwsYNBp+Q61PF43zyDg9Bm9+3NNySn78jMZUUkumqE4Gp7JmFOdP1vc8PpRrzj9+wPinCy8K1PiJ4aYbnTYpCCbDkBSbzhu2QJ1Gd82t8jI8TH51+OzvXoWbnXUOBkNW+0mWFwGcGOUVpU81/n3TOHb5oMt2FgYGjzau0Nif0Ss7Q3XB33hjjQHjHA5E5aOyIQc8CBrLdQSs3j92VG+3nNEjbkbdbBr9zm04ruvw37vh0QKOdeGIkckc80fX3KH/h7PT4BOjgCty8VZ5ux1MoO5Cf5naca2LAsEgehI+drX8o/0Nu+W0m6K/I9gGPd/dfx/EN/wN62AhsBWuAAAAAElFTkSuQmCC
-">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
+        ">
+    </div>
 </div>
 
-## Overview
+# CodeLlama
 
-The Code Llama model was proposed in [Code Llama: Open Foundation Models for Code](https://ai.meta.com/research/publications/code-llama-open-foundation-models-for-code/) by Baptiste Rozière, Jonas Gehring, Fabian Gloeckle, Sten Sootla, Itai Gat, Xiaoqing Ellen Tan, Yossi Adi, Jingyu Liu, Tal Remez, Jérémy Rapin, Artyom Kozhevnikov, Ivan Evtimov, Joanna Bitton, Manish Bhatt, Cristian Canton Ferrer, Aaron Grattafiori, Wenhan Xiong, Alexandre Défossez, Jade Copet, Faisal Azhar, Hugo Touvron, Louis Martin, Nicolas Usunier, Thomas Scialom, Gabriel Synnaeve.
+[Code Llama](https://huggingface.co/papers/2308.12950) is a specialized family of large language models based on [Llama 2](./llama2) for coding tasks.  It comes in different flavors - general code, Python-specific, and instruction-following variant - all available in 7B, 13B, 34B, and 70B parameters. Code Llama models can generate, explain, and even fill in missing parts of your code (called "infilling"). It can also handle very long contexts with stable generation up to 100k tokens, even though it was trained on sequences of 16K tokens.
 
-The abstract from the paper is the following:
+You can find all the original Code Llama checkpoints under the [Code Llama](https://huggingface.co/collections/meta-llama/code-llama-family-661da32d0a9d678b6f55b933) collection.
 
-*We release Code Llama, a family of large language models for code based on Llama 2 providing state-of-the-art performance among open models, infilling capabilities, support for large input contexts, and zero-shot instruction following ability for programming tasks. We provide multiple flavors to cover a wide range of applications: foundation models (Code Llama), Python specializations (Code Llama - Python), and instruction-following models (Code Llama - Instruct) with 7B, 13B and 34B parameters each. All models are trained on sequences of 16k tokens and show improvements on inputs with up to 100k tokens. 7B and 13B Code Llama and Code Llama - Instruct variants support infilling based on surrounding content. Code Llama reaches state-of-the-art performance among open models on several code benchmarks, with scores of up to 53% and 55% on HumanEval and MBPP, respectively. Notably, Code Llama - Python 7B outperforms Llama 2 70B on HumanEval and MBPP, and all our models outperform every other publicly available model on MultiPL-E. We release Code Llama under a permissive license that allows for both research and commercial use.*
+> [!TIP]
+> Click on the Code Llama models in the right sidebar for more examples of how to apply Code Llama to different coding tasks.
 
-Check out all Code Llama model checkpoints [here](https://huggingface.co/models?search=code_llama) and the officially released ones in the [Meta Llama org](https://huggingface.co/meta-llama).
+The example below demonstrates how to generate code with [`Pipeline`], or the [`AutoModel`], and from the command line.
 
-This model was contributed by [ArthurZucker](https://huggingface.co/ArthurZ). The original code of the authors can be found [here](https://github.com/facebookresearch/llama).
+<hfoptions id="usage">
+<hfoption id="Pipeline">
+    
+```py
+import torch
+from transformers import pipeline
 
-## Usage tips and examples
+pipe = pipeline(
+    "text-generation",
+    model="meta-llama/CodeLlama-7b-hf",
+    torch_dtype=torch.float16,
+    device_map=0
+)
 
-<Tip warning={true}>
+# basic code generation
+result = pipe("# Function to calculate the factorial of a number\ndef factorial(n):", max_new_tokens=256)
+print(result[0]['generated_text'])
 
-The `Llama2` family models, on which Code Llama is based, were trained using `bfloat16`, but the original inference uses `float16`. Let's look at the different precisions:
+# infilling
+infill_result = pipe("def remove_non_ascii(s: str) -> str:\n    \"\"\" <FILL_ME>\n    return result", max_new_tokens=200)
+print(infill_result[0]['generated_text'])
+```
 
-* `float32`: PyTorch convention on model initialization is to load models in `float32`, no matter with which `dtype` the model weights were stored. `transformers` also follows this convention for consistency with PyTorch. This will be picked by default. If you want the `AutoModel` API to load the checkpoints with the storage weights type, you must specify `torch_dtype="auto"`, e.g. `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`.
-* `bfloat16`: Code Llama was trained with this precision, so we recommend using it for further training or fine-tuning.
-* `float16`: We recommend running inference using this precision, as it's usually faster than `bfloat16`, and evaluation metrics show no discernible degradation with respect to `bfloat16`. You can also run inference using `bfloat16`, and we recommend you check inference results with both `float16` and `bfloat16` after fine-tuning.
+</hfoption>
+<hfoption id="AutoModel">
 
-As mentioned above, the `dtype` of the storage weights is mostly irrelevant unless you are using `torch_dtype="auto"` when initializing a model using. The reason is that the model will first be downloaded (using the `dtype` of the checkpoints online) and then will be casted to the default `dtype` of `torch` (becomes `torch.float32`). If there is a specified `torch_dtype`, it will be used instead.
+```py
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
-</Tip>
+tokenizer = AutoTokenizer.from_pretrained("meta-llama/CodeLlama-7b-hf")
+model = AutoModelForCausalLM.from_pretrained(
+    "meta-llama/CodeLlama-7b-hf",
+    torch_dtype=torch.float16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
 
+# basic code generation
+prompt = "# Function to calculate the factorial of a number\ndef factorial(n):"
+input_ids = tokenizer(prompt, return_tensors="pt").to("cuda")
 
-Tips:
-- The infilling task is supported out of the box. You should be using the `tokenizer.fill_token` where you want your input to be filled.
-- The model conversion script is the same as for the `Llama2` family:
+output = model.generate(
+    **input_ids, 
+    max_new_tokens=256,
+    cache_implementation="static"
+)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 
-Here is a sample usage:
+# infilling
+infill_prompt = "def remove_non_ascii(s: str) -> str:\n    \"\"\" <FILL_ME>\n    return result"
+input_ids = tokenizer(infill_prompt, return_tensors="pt").to(model.device)
 
+filled_output = model.generate(**input_ids, max_new_tokens=200)
+filled_text = tokenizer.decode(filled_output[0], skip_special_tokens=True)
+print(filled_text)
+```
+
+</hfoption>
+<hfoption id="transformers-cli">
+    
 ```bash
-python src/transformers/models/llama/convert_llama_weights_to_hf.py \
-    --input_dir /path/to/downloaded/llama/weights --model_size 7B --output_dir /output/path
+echo -e "# Function to calculate the factorial of a number\ndef factorial(n):" | transformers-cli run --task text-generation --model meta-llama/CodeLlama-7b-hf --device 0
 ```
 
-Note that executing the script requires enough CPU RAM to host the whole model in float16 precision (even if the biggest versions
-come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM).
+</hfoption>
+</hfoptions>
 
-After conversion, the model and tokenizer can be loaded via:
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
 
-```python
->>> from transformers import LlamaForCausalLM, CodeLlamaTokenizer
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to only quantize the weights to 4-bits.
 
->>> tokenizer = CodeLlamaTokenizer.from_pretrained("meta-llama/CodeLlama-7b-hf")
->>> model = LlamaForCausalLM.from_pretrained("meta-llama/CodeLlama-7b-hf")
->>> PROMPT = '''def remove_non_ascii(s: str) -> str:
-...     """ <FILL_ME>
-...     return result
-... '''
->>> input_ids = tokenizer(PROMPT, return_tensors="pt")["input_ids"]
->>> generated_ids = model.generate(input_ids, max_new_tokens=128)
+```py
+# pip install bitsandbytes
+import torch
+from transformers import AutoModelForCausalLM, CodeLlamaTokenizer, BitsAndBytesConfig
 
->>> filling = tokenizer.batch_decode(generated_ids[:, input_ids.shape[1]:], skip_special_tokens = True)[0]
->>> print(PROMPT.replace("<FILL_ME>", filling))
-def remove_non_ascii(s: str) -> str:
-    """ Remove non-ASCII characters from a string.
-<BLANKLINE>
-    Args:
-        s: The string to remove non-ASCII characters from.
-<BLANKLINE>
-    Returns:
-        The string with non-ASCII characters removed.
-    """
-    result = ""
-    for c in s:
-        if ord(c) < 128:
-            result += c
-    return result
-<BLANKLINE>
+bnb_config = BitsAndBytesConfig(load_in_4bit=True, bnb_4bit_compute_dtype=torch.bfloat16, bnb_4bit_quant_type="nf4", bnb_4bit_use_double_quant=True)
+tokenizer = CodeLlamaTokenizer.from_pretrained("meta-llama/CodeLlama-34b-hf")
+model = AutoModelForCausalLM.from_pretrained(
+   "meta-llama/CodeLlama-34b-hf",
+   torch_dtype=torch.bfloat16,
+   device_map="auto",
+   quantization_config=bnb_config
+)
+
+prompt = "# Write a Python function to check if a string is a palindrome\ndef is_palindrome(s):"
+input_ids = tokenizer(prompt, return_tensors="pt").to("cuda")
+
+output = model.generate(**input_ids, max_new_tokens=200, cache_implementation="static")
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 
-If you only want the infilled part:
-```python
->>> from transformers import pipeline
->>> import torch
+Use the [AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139) to better understand what tokens the model can and cannot attend to.
 
->>> generator = pipeline("text-generation",model="meta-llama/CodeLlama-7b-hf",torch_dtype=torch.float16, device_map="auto")
->>> generator('def remove_non_ascii(s: str) -> str:\n    """ <FILL_ME>\n    return result', max_new_tokens = 128)
-[{'generated_text': 'def remove_non_ascii(s: str) -> str:\n    """ <FILL_ME>\n    return resultRemove non-ASCII characters from a string. """\n    result = ""\n    for c in s:\n        if ord(c) < 128:\n            result += c'}]
+```py
+from transformers.utils.attention_visualizer import AttentionMaskVisualizer
+
+visualizer = AttentionMaskVisualizer("meta-llama/CodeLlama-7b-hf")
+visualizer("""def func(a, b):
+  return a + b""")
 ```
 
-Under the hood, the tokenizer [automatically splits by `<FILL_ME>`](https://huggingface.co/docs/transformers/main/model_doc/code_llama#transformers.CodeLlamaTokenizer.fill_token) to create a formatted input string that follows [the original training pattern](https://github.com/facebookresearch/codellama/blob/cb51c14ec761370ba2e2bc351374a79265d0465e/llama/generation.py#L402). This is more robust than preparing the pattern yourself: it avoids pitfalls, such as token glueing, that are very hard to debug.  To see how much CPU and GPU memory you need for this model or others, try [this calculator](https://huggingface.co/spaces/hf-accelerate/model-memory-usage) which can help determine that value.
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/codellama-attn-mask.png"/>
+</div>
 
-The LLaMA tokenizer is a BPE model based on [sentencepiece](https://github.com/google/sentencepiece). One quirk of sentencepiece is that when decoding a sequence, if the first token is the start of the word (e.g. "Banana"), the tokenizer does not prepend the prefix space to the string.
-
-<Tip>
-
-Code Llama has the same architecture as the `Llama2` models, refer to [Llama2's documentation page](llama2) for the API reference.
-Find Code Llama tokenizer reference below. 
-</Tip>
+## Notes
 
+- Infilling is only available in the 7B and 13B base models, and not in the Python, Instruct, 34B, or 70B models.
+- Use the `<FILL_ME>` token where you want your input to be filled. The tokenizer splits this token to create a formatted input string that follows the [original training pattern](https://github.com/facebookresearch/codellama/blob/cb51c14ec761370ba2e2bc351374a79265d0465e/llama/generation.py#L402). This is more robust than preparing the pattern yourself.
+    ```py
+    from transformers import LlamaForCausalLM, CodeLlamaTokenizer
+    
+    tokenizer = CodeLlamaTokenizer.from_pretrained("meta-llama/CodeLlama-7b-hf")
+    model = LlamaForCausalLM.from_pretrained("meta-llama/CodeLlama-7b-hf")
+    PROMPT = '''def remove_non_ascii(s: str) -> str:
+        """ <FILL_ME>
+        return result
+    '''
+    input_ids = tokenizer(PROMPT, return_tensors="pt")["input_ids"]
+    generated_ids = model.generate(input_ids, max_new_tokens=128)
+    
+    filling = tokenizer.batch_decode(generated_ids[:, input_ids.shape[1]:], skip_special_tokens = True)[0]
+    print(PROMPT.replace("<FILL_ME>", filling))
+    ```
+- Use `bfloat16` for further training or fine-tuning and `float16` for inference.
+- The `BOS` character is not used for infilling when encoding the prefix or suffix, but only at the beginning of each prompt.
+- The tokenizer is a byte-pair encoding model based on [SentencePiece](https://github.com/google/sentencepiece). During decoding, if the first token is the start of the word (for example, “Banana”), the tokenizer doesn’t prepend the prefix space to the string.
 
 ## CodeLlamaTokenizer
 

docs/source/en/model_doc/cohere.md

@@ -1,124 +1,115 @@
-# Cohere
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
 
-The Cohere Command-R model was proposed in the blogpost [Command-R: Retrieval Augmented Generation at Production Scale](https://txt.cohere.com/command-r/) by the Cohere Team.
+# Cohere
 
-The abstract from the paper is the following:
+Cohere Command-R is a 35B parameter multilingual large language model designed for long context tasks like retrieval-augmented generation (RAG) and calling external APIs and tools. The model is specifically trained for grounded generation and supports both single-step and multi-step tool use. It supports a context length of 128K tokens.
 
-*Command-R is a scalable generative model targeting RAG and Tool Use to enable production-scale AI for enterprise. Today, we are introducing Command-R, a new LLM aimed at large-scale production workloads. Command-R targets the emerging “scalable” category of models that balance high efficiency with strong accuracy, enabling companies to move beyond proof of concept, and into production.*
+You can find all the original Command-R checkpoints under the [Command Models](https://huggingface.co/collections/CohereForAI/command-models-67652b401665205e17b192ad) collection.
 
-*Command-R is a generative model optimized for long context tasks such as retrieval augmented generation (RAG) and using external APIs and tools. It is designed to work in concert with our industry-leading Embed and Rerank models to provide best-in-class integration for RAG applications and excel at enterprise use cases. As a model built for companies to implement at scale, Command-R boasts:
-- Strong accuracy on RAG and Tool Use
-- Low latency, and high throughput
-- Longer 128k context and lower pricing
-- Strong capabilities across 10 key languages
-- Model weights available on HuggingFace for research and evaluation
 
-Checkout model checkpoints [here](https://huggingface.co/CohereForAI/c4ai-command-r-v01).
-This model was contributed by [Saurabh Dash](https://huggingface.co/saurabhdash) and [Ahmet Üstün](https://huggingface.co/ahmetustun). The code of the implementation in Hugging Face is based on GPT-NeoX [here](https://github.com/EleutherAI/gpt-neox).
+> [!TIP]
+> Click on the Cohere models in the right sidebar for more examples of how to apply Cohere to different language tasks.
 
-## Usage tips
+The example below demonstrates how to generate text with [`Pipeline`] or the [`AutoModel`], and from the command line.
 
-<Tip warning={true}>
-
-The checkpoints uploaded on the Hub use `torch_dtype = 'float16'`, which will be
-used by the `AutoModel` API to cast the checkpoints from `torch.float32` to `torch.float16`. 
-
-The `dtype` of the online weights is mostly irrelevant unless you are using `torch_dtype="auto"` when initializing a model using `model = AutoModelForCausalLM.from_pretrained("path", torch_dtype = "auto")`. The reason is that the model will first be downloaded ( using the `dtype` of the checkpoints online), then it will be casted to the default `dtype` of `torch` (becomes `torch.float32`), and finally, if there is a `torch_dtype` provided in the config, it will be used. 
-
-Training the model in `float16` is not recommended and is known to produce `nan`; as such, the model should be trained in `bfloat16`.
-
-</Tip>
-The model and tokenizer can be loaded via:
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
 ```python
-# pip install transformers
+import torch
+from transformers import pipeline
+
+pipeline = pipeline(
+    task="text-generation",
+    model="CohereForAI/c4ai-command-r-v01",
+    torch_dtype=torch.float16,
+    device=0
+)
+pipeline("Plants create energy through a process known as")
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```python
+import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 
-model_id = "CohereForAI/c4ai-command-r-v01"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
+tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-v01")
+model = AutoModelForCausalLM.from_pretrained("CohereForAI/c4ai-command-r-v01", torch_dtype=torch.float16, device_map="auto", attn_implementation="sdpa")
 
-# Format message with the command-r chat template
-messages = [{"role": "user", "content": "Hello, how are you?"}]
-input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
-## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
-
-gen_tokens = model.generate(
+# format message with the Command-R chat template
+messages = [{"role": "user", "content": "How do plants make energy?"}]
+input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
+output = model.generate(
     input_ids, 
     max_new_tokens=100, 
     do_sample=True, 
     temperature=0.3,
-    )
-
-gen_text = tokenizer.decode(gen_tokens[0])
-print(gen_text)
+    cache_implementation="static",
+)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 
-- When using Flash Attention 2 via `attn_implementation="flash_attention_2"`, don't pass `torch_dtype` to the `from_pretrained` class method and use Automatic Mixed-Precision training. When using `Trainer`, it is simply specifying either `fp16` or `bf16` to `True`. Otherwise, make sure you are using `torch.autocast`. This is required because the Flash Attention only support `fp16` and `bf16` data type.
+</hfoption>
+<hfoption id="transformers-cli">
 
-
-## Resources
-
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Command-R. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
-
-
-<PipelineTag pipeline="text-generation"/>
-
-Loading FP16 model
-```python
-# pip install transformers
-from transformers import AutoTokenizer, AutoModelForCausalLM
-
-model_id = "CohereForAI/c4ai-command-r-v01"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
-
-# Format message with the command-r chat template
-messages = [{"role": "user", "content": "Hello, how are you?"}]
-input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt")
-## <BOS_TOKEN><|START_OF_TURN_TOKEN|><|USER_TOKEN|>Hello, how are you?<|END_OF_TURN_TOKEN|><|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>
-
-gen_tokens = model.generate(
-    input_ids, 
-    max_new_tokens=100, 
-    do_sample=True, 
-    temperature=0.3,
-    )
-
-gen_text = tokenizer.decode(gen_tokens[0])
-print(gen_text)
+```bash
+# pip install -U flash-attn --no-build-isolation
+transformers-cli chat --model_name_or_path CohereForAI/c4ai-command-r-v01 --torch_dtype auto --attn_implementation flash_attention_2
 ```
 
-Loading bitsnbytes 4bit quantized model
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [bitsandbytes](../quantization/bitsandbytes) to quantize the weights to 4-bits.
+
 ```python
-# pip install transformers bitsandbytes accelerate
-from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
+import torch
+from transformers import BitsAndBytesConfig, AutoTokenizer, AutoModelForCausalLM
 
 bnb_config = BitsAndBytesConfig(load_in_4bit=True)
+tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-v01")
+model = AutoModelForCausalLM.from_pretrained("CohereForAI/c4ai-command-r-v01", torch_dtype=torch.float16, device_map="auto", quantization_config=bnb_config, attn_implementation="sdpa")
 
-model_id = "CohereForAI/c4ai-command-r-v01"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=bnb_config)
-
-gen_tokens = model.generate(
+# format message with the Command-R chat template
+messages = [{"role": "user", "content": "How do plants make energy?"}]
+input_ids = tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True, return_tensors="pt").to("cuda")
+output = model.generate(
     input_ids, 
     max_new_tokens=100, 
     do_sample=True, 
     temperature=0.3,
-    )
-
-gen_text = tokenizer.decode(gen_tokens[0])
-print(gen_text)
+    cache_implementation="static",
+)
+print(tokenizer.decode(output[0], skip_special_tokens=True))
 ```
 
+Use the [AttentionMaskVisualizer](https://github.com/huggingface/transformers/blob/beb9b5b02246b9b7ee81ddf938f93f44cfeaad19/src/transformers/utils/attention_visualizer.py#L139) to better understand what tokens the model can and cannot attend to.
+
+```py
+from transformers.utils.attention_visualizer import AttentionMaskVisualizer
+
+visualizer = AttentionMaskVisualizer("CohereForAI/c4ai-command-r-v01")
+visualizer("Plants create energy through a process known as")
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/cohere-attn-mask.png"/>
+</div>
+
+
+## Notes
+- Don’t use the torch_dtype parameter in [`~AutoModel.from_pretrained`] if you’re using FlashAttention-2 because it only supports fp16 or bf16. You should use [Automatic Mixed Precision](https://pytorch.org/tutorials/recipes/recipes/amp_recipe.html), set fp16 or bf16 to True if using [`Trainer`], or use [torch.autocast](https://pytorch.org/docs/stable/amp.html#torch.autocast).
 
 ## CohereConfig
 
@@ -143,5 +134,3 @@ print(gen_text)
 
 [[autodoc]] CohereForCausalLM
     - forward
-
-

docs/source/en/model_doc/colpali.md

@@ -1,5 +1,4 @@
 <!--Copyright 2024 The HuggingFace Team. All rights reserved.
-
 Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
 the License. You may obtain a copy of the License at
 
@@ -9,76 +8,134 @@ Unless required by applicable law or agreed to in writing, software distributed
 an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
 specific language governing permissions and limitations under the License.
 
-⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+⚠️ Note that this file is in Markdown but contains specific syntax for our doc-builder (similar to MDX) that may not be
 rendered properly in your Markdown viewer.
-
 -->
 
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
+
 # ColPali
 
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-</div>
+[ColPali](https://huggingface.co/papers/2407.01449) is a model designed to retrieve documents by analyzing their visual features. Unlike traditional systems that rely heavily on text extraction and OCR, ColPali treats each page as an image. It uses [Paligemma-3B](./paligemma) to capture not only text, but also the layout, tables, charts, and other visual elements to create detailed embeddings. This offers a more comprehensive understanding of documents and enables more efficient and accurate retrieval.
 
-## Overview
+You can find all the original ColPali checkpoints under the [ColPali](https://huggingface.co/collections/vidore/hf-native-colvision-models-6755d68fc60a8553acaa96f7) collection.
 
-The *ColPali* model was proposed in [ColPali: Efficient Document Retrieval with Vision Language Models](https://doi.org/10.48550/arXiv.2407.01449) by **Manuel Faysse***, **Hugues Sibille***, **Tony Wu***, Bilel Omrani, Gautier Viaud, Céline Hudelot, Pierre Colombo (* denotes equal contribution). Work lead by ILLUIN Technology.
+> [!TIP]
+> Click on the ColPali models in the right sidebar for more examples of how to use ColPali for image retrieval.
 
-In our proposed *ColPali* approach, we leverage VLMs to construct efficient multi-vector embeddings directly from document images (“screenshots”) for document retrieval. We train the model to maximize the similarity between these document embeddings and the corresponding query embeddings, using the late interaction method introduced in ColBERT.
+<hfoptions id="usage">
+<hfoption id="image retrieval">
 
-Using *ColPali* removes the need for potentially complex and brittle layout recognition and OCR pipelines with a single model that can take into account both the textual and visual content (layout, charts, etc.) of a document.
-
-## Resources
-
-- The *ColPali* arXiv paper can be found [here](https://doi.org/10.48550/arXiv.2407.01449). 📄
-- The official blog post detailing ColPali can be found [here](https://huggingface.co/blog/manu/colpali). 📝
-- The original model implementation code for the ColPali model and for the `colpali-engine` package can be found [here](https://github.com/illuin-tech/colpali). 🌎
-- Cookbooks for learning to use the transformers-native version of *ColPali*, fine-tuning, and similarity maps generation can be found [here](https://github.com/tonywu71/colpali-cookbooks). 📚
-
-This model was contributed by [@tonywu71](https://huggingface.co/tonywu71) and [@yonigozlan](https://huggingface.co/yonigozlan).
-
-## Usage
-
-This example demonstrates how to use *ColPali* to embed both queries and images, calculate their similarity scores, and identify the most relevant matches. For a specific query, you can retrieve the top-k most similar images by selecting the ones with the highest similarity scores.
-
-```python
+```py
+import requests
 import torch
 from PIL import Image
-
 from transformers import ColPaliForRetrieval, ColPaliProcessor
 
-model_name = "vidore/colpali-v1.2-hf"
-
+# Load model (bfloat16 support is limited; fallback to float32 if needed)
 model = ColPaliForRetrieval.from_pretrained(
-    model_name,
-    torch_dtype=torch.bfloat16,
-    device_map="cuda:0",  # or "mps" if on Apple Silicon
+    "vidore/colpali-v1.2-hf",
+    torch_dtype=torch.bfloat16 if torch.cuda.is_available() else torch.float32,
+    device_map="auto",  # "cpu", "cuda", or "mps" for Apple Silicon
 ).eval()
 
 processor = ColPaliProcessor.from_pretrained(model_name)
 
-# Your inputs (replace dummy images with screenshots of your documents)
+url1 = "https://upload.wikimedia.org/wikipedia/commons/8/89/US-original-Declaration-1776.jpg"
+url2 = "https://upload.wikimedia.org/wikipedia/commons/thumb/4/4c/Romeoandjuliet1597.jpg/500px-Romeoandjuliet1597.jpg"
+
 images = [
-    Image.new("RGB", (32, 32), color="white"),
-    Image.new("RGB", (16, 16), color="black"),
+    Image.open(requests.get(url1, stream=True).raw),
+    Image.open(requests.get(url2, stream=True).raw),
 ]
+
 queries = [
-    "What is the organizational structure for our R&D department?",
-    "Can you provide a breakdown of last year’s financial performance?",
+    "Who printed the edition of Romeo and Juliet?",
+    "When was the United States Declaration of Independence proclaimed?",
 ]
 
 # Process the inputs
-batch_images = processor(images=images).to(model.device)
-batch_queries = processor(text=queries).to(model.device)
+inputs_images = processor(images=images, return_tensors="pt").to(model.device)
+inputs_text = processor(text=queries, return_tensors="pt").to(model.device)
 
 # Forward pass
 with torch.no_grad():
-    image_embeddings = model(**batch_images).embeddings
-    query_embeddings = model(**batch_queries).embeddings
+    image_embeddings = model(**inputs_images).embeddings
+    query_embeddings = model(**inputs_text).embeddings
 
-# Score the queries against the images
 scores = processor.score_retrieval(query_embeddings, image_embeddings)
+
+print("Retrieval scores (query x image):")
+print(scores)
 ```
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [bitsandbytes](../quantization/bitsandbytes.md) to quantize the weights to int4.
+
+```py
+import requests
+import torch
+from PIL import Image
+from transformers import ColPaliForRetrieval, ColPaliProcessor
+from transformers import BitsAndBytesConfig
+
+# 4-bit quantization configuration
+bnb_config = BitsAndBytesConfig(
+    load_in_4bit=True,
+    bnb_4bit_use_double_quant=True,
+    bnb_4bit_quant_type="nf4",
+    bnb_4bit_compute_dtype=torch.float16,
+)
+
+model_name = "vidore/colpali-v1.2-hf"
+
+# Load model 
+model = ColPaliForRetrieval.from_pretrained(
+    model_name,
+    quantization_config=bnb_config,
+    device_map="cuda"
+).eval()
+
+processor = ColPaliProcessor.from_pretrained(model_name)
+
+url1 = "https://upload.wikimedia.org/wikipedia/commons/8/89/US-original-Declaration-1776.jpg"
+url2 = "https://upload.wikimedia.org/wikipedia/commons/thumb/4/4c/Romeoandjuliet1597.jpg/500px-Romeoandjuliet1597.jpg"
+
+images = [
+    Image.open(requests.get(url1, stream=True).raw),
+    Image.open(requests.get(url2, stream=True).raw),
+]
+
+queries = [
+    "Who printed the edition of Romeo and Juliet?",
+    "When was the United States Declaration of Independence proclaimed?",
+]
+
+# Process the inputs
+inputs_images = processor(images=images, return_tensors="pt").to(model.device)
+inputs_text = processor(text=queries, return_tensors="pt").to(model.device)
+
+# Forward pass
+with torch.no_grad():
+    image_embeddings = model(**inputs_images).embeddings
+    query_embeddings = model(**inputs_text).embeddings
+
+scores = processor.score_retrieval(query_embeddings, image_embeddings)
+
+print("Retrieval scores (query x image):")
+print(scores)
+```
+
+## Notes
+
+- [`~ColPaliProcessor.score_retrieval`] returns a 2D tensor where the first dimension is the number of queries and the second dimension is the number of images. A higher score indicates more similarity between the query and image.
 
 ## ColPaliConfig
 

docs/source/en/model_doc/conditional_detr.md

@@ -48,6 +48,11 @@ This model was contributed by [DepuMeng](https://huggingface.co/DepuMeng). The o
 
 [[autodoc]] ConditionalDetrImageProcessor
     - preprocess
+
+## ConditionalDetrImageProcessorFast
+
+[[autodoc]] ConditionalDetrImageProcessorFast
+    - preprocess
     - post_process_object_detection
     - post_process_instance_segmentation
     - post_process_semantic_segmentation

docs/source/en/model_doc/deepseek_v3.md

@@ -0,0 +1,184 @@
+<!--Copyright 2025 The HuggingFace Team. All rights reserved.
+
+Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with
+the License. You may obtain a copy of the License at
+
+http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on
+an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the
+specific language governing permissions and limitations under the License.
+
+⚠️ Note that this file is in Markdown but contain specific syntax for our doc-builder (similar to MDX) that may not be
+rendered properly in your Markdown viewer.
+
+-->
+
+# DeepSeek-V3
+
+## Overview
+
+The DeepSeek-V3 model was proposed in [DeepSeek-V3 Technical Report](https://arxiv.org/abs/2412.19437) by DeepSeek-AI Team.
+
+The abstract from the paper is the following:
+We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations reveal that DeepSeek-V3 outperforms other open-source models and achieves performance comparable to leading closed-source models. Despite its excellent performance, DeepSeek-V3 requires only 2.788M H800 GPU hours for its full training. In addition, its training process is remarkably stable. Throughout the entire training process, we did not experience any irrecoverable loss spikes or perform any rollbacks. The model checkpoints are available at https://github.com/deepseek-ai/DeepSeek-V3.
+
+## Limitations and call for contribution!
+
+We are super happy to make this code community-powered, and would love to see how you can best optimize the following: 
+
+- current implementation uses the "naive" attention compution (so not really MLA)
+- current implementation loops through the experts. This should be replaced. Pointers to use `get_packed_weights` from `intetrations/tensor_parallel`. 
+- current implementation uses the eleuther formula for ROPE, using the orginal one would be more efficient! (should still follow our API)
+- static cache is not supported (this should be just a generation config issue / config shape issues)
+
+### Usage tips
+The model uses Multi-head Latent Attention (MLA) and DeepSeekMoE architectures for efficient inference and cost-effective training. It employs an auxiliary-loss-free strategy for load balancing and multi-token prediction training objective. The model can be used for various language tasks after being pre-trained on 14.8 trillion tokens and going through Supervised Fine-Tuning and Reinforcement Learning stages.
+
+You can run the model in `FP8` automatically, using 2 nodes of 8 H100 should be more than enough! 
+
+```python
+# `run_deepseek_v1.py`
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+torch.manual_seed(30)
+
+tokenizer = AutoTokenizer.from_pretrained("deepseek-r1")
+
+chat = [
+  {"role": "user", "content": "Hello, how are you?"},
+  {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
+  {"role": "user", "content": "I'd like to show off how chat templating works!"},
+]
+
+
+model = AutoModelForCausalLM.from_pretrained("deepseek-r1", device_map="auto", torch_dtype=torch.bfloat16)
+inputs = tokenizer.apply_chat_template(chat, tokenize=True, add_generation_prompt=True, return_tensors="pt").to(model.device)
+import time
+start = time.time()
+outputs = model.generate(inputs, max_new_tokens=50)
+print(tokenizer.batch_decode(outputs))
+print(time.time()-start)
+```
+This generated: 
+
+``````
+<｜Assistant｜><think>
+Okay, the user wants to demonstrate how chat templating works. Let me break down what that means. Chat templating is about structuring the conversation data, especially for models that need specific input formats. Maybe they're referring to something like how messages are formatted with roles (user, assistant, system) in APIs like OpenAI.
+
+First, I should explain what chat templating is. It's the process of formatting conversation data into a structured format that the model can understand. This usually includes roles and content. For example, user messages, assistant responses, and system messages each have their own role tags.
+
+They might want an example. Let me think of a simple conversation. The user says "Hello, how are you?" and the assistant responds "I'm doing great. How can I help you today?" Then the user follows up with wanting to show off chat templating. So the example should include the history and the new message.
+
+In some frameworks, like Hugging Face's Transformers, chat templates are applied using Jinja2 templates. The template might look something like combining system messages, then looping through user and assistant messages with appropriate tags. For instance, using {% for message in messages %} and assigning roles like <|user|>, <|assistant|>, etc.
+
+I should structure the example with the messages array, showing each role and content. Then apply a hypothetical template to convert that into a formatted string the model uses. Also, mention that different models have different templating requirements, like using special tokens or varying role labels.
+
+Wait, the user mentioned "chat templating" in the context of showing off. Maybe they want a practical example they can present. So providing a code snippet or a structured data example would be helpful. Let me outline a typical messages array and then the templated output.
+
+Also, it's important to note that proper templating ensures the model knows the conversation flow, which is crucial for generating coherent responses. Maybe include a note about why it's important, like maintaining context and role-specific processing.
+
+Let me check if there are any common mistakes or things to avoid. For example, not closing tags properly, or mismatching roles. But maybe that's too detailed unless the user asks. Focus on the positive example first.
+
+Putting it all together, the response should have an example messages array, the applied template, and the final formatted string. Maybe use angle brackets or special tokens as placeholders. Also, mention that this helps in training or fine-tuning models with structured data.
+
+I think that's a solid approach. Let me structure it step by step to make it clear.
+</think>
+
+Chat templating is a way to structure conversation data (e.g., user/assistant interactions) into a format that language models understand. This is especially important for models trained to handle multi-turn dialogues, where the input must explicitly separate roles (user, assistant, system, etc.) and messages. Let’s break this down with an example!
+
+---
+
+### **Step 1: Raw Conversation History**
+Suppose we have this conversation:
+- **User**: "Hello, how are you?"
+- **Assistant**: "I'm doing great. How can I help you today?"
+- **User**: "I'd like to show off how chat templating works!"
+
+---
+
+### **Step 2: Structured Messages**
+In frameworks like Hugging Face Transformers or OpenAI, conversations are often formatted as a list of dictionaries with `role` and `content`:
+```python
+messages = [
+    {"role": "user", "content": "Hello, how are you?"},
+    {"role": "assistant", "content": "I'm doing great. How can I help you today?"},
+    {"role": "user", "content": "I'd like to show off how chat templating works!"},
+]
+```
+
+---
+
+### **Step 3: Apply a Chat Template**
+A **chat template** converts this structured data into a single string formatted for the model. For example, using a Jinja-style template (common in Hugging Face):
+
+```jinja
+{% for message in messages %}
+    {% if message['role'] == 'user' %}
+        <|user|>{{ message['content'] }}<|end|>
+    {% elif message['role'] == 'assistant' %}
+        <|assistant|>{{ message['content'] }}<|end|>
+    {% endif %}
+{% endfor %}
+<|assistant|>
+```
+
+---
+
+### **Step 4: Final Templated Output**
+Applying the template to our `messages` list would produce:
+```text
+<|user|>Hello, how are you?<|end|>
+<|assistant|>I'm doing great. How can I help you today?<|end|>
+<|user|>I'd like to show off how chat templating works!<|end|>
+<|assistant|>
+```
+
+This tells the model:  
+1. The conversation history (user/assistant turns).  
+2. The model’s turn to generate a response (`<|assistant|>` at the end).  
+
+---
+
+### **Key Notes**:
+- **Role Separation**: Tags like `<|user|>` and `<|assistant|>` help the model distinguish speakers.
+- **Special Tokens**: Models often use unique tokens (e.g., `<|end|>`) to mark message boundaries.
+- **Flexibility**: Templates vary by model (e.g., OpenAI uses `{"role": "user", "content": "..."}` instead of tags).
+
+---
+
+### **Why This Matters**:
+- **Consistency**: Ensures the model understands dialogue structure.
+- **Context Preservation**: Maintains the flow of multi-turn conversations.
+- **Alignment**: Matches the format the model was trained on for better performance.
+
+Want to dive deeper or see a specific framework’s implementation (e.g., OpenAI, Llama, Mistral)? Let me know! 😊<｜end▁of▁sentence｜>
+``````
+
+Use the following to run it
+```bash
+torchrun --nproc_per_node=8 --nnodes=2 --node_rank=0|1 --rdzv-id an_id --rdzv-backend c10d --rdzv-endpoint master_addr:master_port run_deepseek_r1.py
+```
+
+If you have: 
+```bash
+[rank0]: ncclInternalError: Internal check failed.
+[rank0]: Last error:
+[rank0]: Bootstrap : no socket interface found
+```
+error, it means NCCL was probably not loaded. 
+
+
+## DeepseekV3Config
+
+[[autodoc]] DeepseekV3Config
+
+## DeepseekV3Model
+
+[[autodoc]] DeepseekV3Model
+    - forward
+
+## DeepseekV3ForCausalLM
+
+[[autodoc]] DeepseekV3ForCausalLM
+    - forward

docs/source/en/model_doc/deit.md

@@ -19,6 +19,7 @@ rendered properly in your Markdown viewer.
 <div class="flex flex-wrap space-x-1">
 <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
 <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
 <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
 </div>
 

docs/source/en/model_doc/depth_anything.md

@@ -14,101 +14,69 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# Depth Anything
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
+# Depth Anything
 
-The Depth Anything model was proposed in [Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data](https://arxiv.org/abs/2401.10891) by Lihe Yang, Bingyi Kang, Zilong Huang, Xiaogang Xu, Jiashi Feng, Hengshuang Zhao. Depth Anything is based on the [DPT](dpt) architecture, trained on ~62 million images, obtaining state-of-the-art results for both relative and absolute depth estimation.
+[Depth Anything](https://huggingface.co/papers/2401.10891) is designed to be a foundation model for monocular depth estimation (MDE). It is jointly trained on labeled and ~62M unlabeled images to enhance the dataset. It uses a pretrained [DINOv2](./dinov2) model as an image encoder to inherit its existing rich semantic priors, and [DPT](./dpt) as the decoder. A teacher model is trained on unlabeled images to create pseudo-labels. The student model is trained on a combination of the pseudo-labels and labeled images. To improve the student model's performance, strong perturbations are added to the unlabeled images to challenge the student model to learn more visual knowledge from the image.
 
-<Tip>
+You can find all the original Depth Anything checkpoints under the [Depth Anything](https://huggingface.co/collections/LiheYoung/depth-anything-release-65b317de04eec72abf6b55aa) collection.
 
-[Depth Anything V2](depth_anything_v2) was released in June 2024. It uses the same architecture as Depth Anything and therefore it is compatible with all code examples and existing workflows. However, it leverages synthetic data and a larger capacity teacher model to achieve much finer and robust depth predictions.
+> [!TIP]
+> Click on the Depth Anything models in the right sidebar for more examples of how to apply Depth Anything to different vision tasks.
 
-</Tip>
+The example below demonstrates how to obtain a depth map with [`Pipeline`] or the [`AutoModel`] class.
 
-The abstract from the paper is the following:
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-*This work presents Depth Anything, a highly practical solution for robust monocular depth estimation. Without pursuing novel technical modules, we aim to build a simple yet powerful foundation model dealing with any images under any circumstances. To this end, we scale up the dataset by designing a data engine to collect and automatically annotate large-scale unlabeled data (~62M), which significantly enlarges the data coverage and thus is able to reduce the generalization error. We investigate two simple yet effective strategies that make data scaling-up promising. First, a more challenging optimization target is created by leveraging data augmentation tools. It compels the model to actively seek extra visual knowledge and acquire robust representations. Second, an auxiliary supervision is developed to enforce the model to inherit rich semantic priors from pre-trained encoders. We evaluate its zero-shot capabilities extensively, including six public datasets and randomly captured photos. It demonstrates impressive generalization ability. Further, through fine-tuning it with metric depth information from NYUv2 and KITTI, new SOTAs are set. Our better depth model also results in a better depth-conditioned ControlNet.*
+```py
+import torch
+from transformers import pipeline
 
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/depth_anything_overview.jpg"
-alt="drawing" width="600"/>
-
-<small> Depth Anything overview. Taken from the <a href="https://arxiv.org/abs/2401.10891">original paper</a>.</small>
-
-This model was contributed by [nielsr](https://huggingface.co/nielsr).
-The original code can be found [here](https://github.com/LiheYoung/Depth-Anything).
-
-## Usage example
-
-There are 2 main ways to use Depth Anything: either using the pipeline API, which abstracts away all the complexity for you, or by using the `DepthAnythingForDepthEstimation` class yourself.
-
-### Pipeline API
-
-The pipeline allows to use the model in a few lines of code:
-
-```python
->>> from transformers import pipeline
->>> from PIL import Image
->>> import requests
-
->>> # load pipe
->>> pipe = pipeline(task="depth-estimation", model="LiheYoung/depth-anything-small-hf")
-
->>> # load image
->>> url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
->>> image = Image.open(requests.get(url, stream=True).raw)
-
->>> # inference
->>> depth = pipe(image)["depth"]
+pipe = pipeline(task="depth-estimation", model="LiheYoung/depth-anything-base-hf", torch_dtype=torch.bfloat16, device=0)
+pipe("http://images.cocodataset.org/val2017/000000039769.jpg")["depth"]
 ```
 
-### Using the model yourself
+</hfoption>
+<hfoption id="AutoModel">
 
-If you want to do the pre- and postprocessing yourself, here's how to do that:
+```py
+import torch
+import requests
+import numpy as np
+from PIL import Image
+from transformers import AutoImageProcessor, AutoModelForDepthEstimation
 
-```python
->>> from transformers import AutoImageProcessor, AutoModelForDepthEstimation
->>> import torch
->>> import numpy as np
->>> from PIL import Image
->>> import requests
+image_processor = AutoImageProcessor.from_pretrained("LiheYoung/depth-anything-base-hf")
+model = AutoModelForDepthEstimation.from_pretrained("LiheYoung/depth-anything-base-hf", torch_dtype=torch.bfloat16)
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+inputs = image_processor(images=image, return_tensors="pt")
 
->>> url = "http://images.cocodataset.org/val2017/000000039769.jpg"
->>> image = Image.open(requests.get(url, stream=True).raw)
+with torch.no_grad():
+    outputs = model(**inputs)
 
->>> image_processor = AutoImageProcessor.from_pretrained("LiheYoung/depth-anything-small-hf")
->>> model = AutoModelForDepthEstimation.from_pretrained("LiheYoung/depth-anything-small-hf")
-
->>> # prepare image for the model
->>> inputs = image_processor(images=image, return_tensors="pt")
-
->>> with torch.no_grad():
-...     outputs = model(**inputs)
-
->>> # interpolate to original size and visualize the prediction
->>> post_processed_output = image_processor.post_process_depth_estimation(
-...     outputs,
-...     target_sizes=[(image.height, image.width)],
-... )
-
->>> predicted_depth = post_processed_output[0]["predicted_depth"]
->>> depth = (predicted_depth - predicted_depth.min()) / (predicted_depth.max() - predicted_depth.min())
->>> depth = depth.detach().cpu().numpy() * 255
->>> depth = Image.fromarray(depth.astype("uint8"))
+post_processed_output = image_processor.post_process_depth_estimation(
+    outputs,
+    target_sizes=[(image.height, image.width)],
+)
+predicted_depth = post_processed_output[0]["predicted_depth"]
+depth = (predicted_depth - predicted_depth.min()) / (predicted_depth.max() - predicted_depth.min())
+depth = depth.detach().cpu().numpy() * 255
+Image.fromarray(depth.astype("uint8"))
 ```
 
-## Resources
+</hfoption>
+</hfoptions>
 
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with Depth Anything.
+## Notes
 
-- [Monocular depth estimation task guide](../tasks/monocular_depth_estimation)
-- A notebook showcasing inference with [`DepthAnythingForDepthEstimation`] can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/blob/master/Depth%20Anything/Predicting_depth_in_an_image_with_Depth_Anything.ipynb). 🌎
-
-If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
+- [DepthAnythingV2](./depth_anything_v2), released in June 2024, uses the same architecture as Depth Anything and is compatible with all code examples and existing workflows. It uses synthetic data and a larger capacity teacher model to achieve much finer and robust depth predictions.
 
 ## DepthAnythingConfig
 

docs/source/en/model_doc/depth_pro.md

@@ -90,7 +90,7 @@ The `DepthProEncoder` further uses two encoders:
 - `image_encoder`
    - Input image is also rescaled to `patch_size` and processed by the **`image_encoder`**
 
-Both these encoders can be configured via `patch_model_config` and `image_model_config` respectively, both of which are seperate `Dinov2Model` by default.
+Both these encoders can be configured via `patch_model_config` and `image_model_config` respectively, both of which are separate `Dinov2Model` by default.
 
 Outputs from both encoders (`last_hidden_state`) and selected intermediate states (`hidden_states`) from **`patch_encoder`** are fused by a `DPT`-based `FeatureFusionStage` for depth estimation.
 

docs/source/en/model_doc/dinov2.md

@@ -10,70 +10,134 @@ an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express o
 specific language governing permissions and limitations under the License.
 -->
 
-# DINOv2
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAC0AAAAtCAMAAAANxBKoAAAC7lBMVEUAAADg5vYHPVgAoJH+/v76+v39/f9JbLP///9+AIgAnY3///+mcqzt8fXy9fgkXa3Ax9709fr+///9/f8qXq49qp5AaLGMwrv8/P0eW60VWawxYq8yqJzG2dytt9Wyu9elzci519Lf3O3S2efY3OrY0+Xp7PT///////+dqNCexMc6Z7AGpJeGvbenstPZ5ejQ1OfJzOLa7ejh4+/r8fT29vpccbklWK8PVa0AS6ghW63O498vYa+lsdKz1NDRt9Kw1c672tbD3tnAxt7R6OHp5vDe7OrDyuDn6vLl6/EAQKak0MgATakkppo3ZK/Bz9y8w9yzu9jey97axdvHzeG21NHH4trTwthKZrVGZLSUSpuPQJiGAI+GAI8SWKydycLL4d7f2OTi1+S9xNzL0ePT6OLGzeEAo5U0qJw/aLEAo5JFa7JBabEAp5Y4qZ2QxLyKmsm3kL2xoMOehrRNb7RIbbOZgrGre68AUqwAqZqNN5aKJ5N/lMq+qsd8kMa4pcWzh7muhLMEV69juq2kbKqgUaOTR5uMMZWLLZSGAI5VAIdEAH+ovNDHuNCnxcy3qcaYx8K8msGplrx+wLahjbYdXrV6vbMvYK9DrZ8QrZ8tqJuFms+Sos6sw8ecy8RffsNVeMCvmb43aLltv7Q4Y7EZWK4QWa1gt6meZKUdr6GOAZVeA4xPAISyveLUwtivxtKTpNJ2jcqfvcltiMiwwcfAoMVxhL+Kx7xjdrqTe60tsaNQs6KaRKACrJ6UTZwkqpqTL5pkHY4AloSgsd2ptNXPvNOOncuxxsqFl8lmg8apt8FJcr9EbryGxLqlkrkrY7dRa7ZGZLQ5t6iXUZ6PPpgVpZeJCJFKAIGareTa0+KJod3H0deY2M+esM25usmYu8d2zsJOdcBVvrCLbqcAOaaHaKQAMaScWqKBXqCXMJ2RHpiLF5NmJZAdAHN2kta11dKu1M+DkcZLdb+Mcql3TppyRJdzQ5ZtNZNlIY+DF4+voCOQAAAAZ3RSTlMABAT+MEEJ/RH+/TP+Zlv+pUo6Ifz8+fco/fz6+evr39S9nJmOilQaF/7+/f38+smmoYp6b1T+/v7++vj189zU0tDJxsGzsrKSfv34+Pf27dDOysG9t6+n/vv6+vr59uzr1tG+tZ6Qg9Ym3QAABR5JREFUSMeNlVVUG1EQhpcuxEspXqS0SKEtxQp1d3d332STTRpIQhIISQgJhODu7lAoDoUCpe7u7u7+1puGpqnCPOyZvffbOXPm/PsP9JfQgyCC+tmTABTOcbxDz/heENS7/1F+9nhvkHePG0wNDLbGWwdXL+rbLWvpmZHXD8+gMfBjTh+aSe6Gnn7lwQIOTR0c8wfX3PWgv7avbdKwf/ZoBp1Gp/PvuvXW3vw5ib7emnTW4OR+3D4jB9vjNJ/7gNvfWWeH/TO/JyYrsiKCRjVEZA3UB+96kON+DxOQ/NLE8PE5iUYgIXjFnCOlxEQMaSGVxjg4gxOnEycGz8bptuNjVx08LscIgrzH3umcn+KKtiBIyvzOO2O99aAdR8cF19oZalnCtvREUw79tCd5sow1g1UKM6kXqUx4T8wsi3sTjJ3yzDmmhenLXLpo8u45eG5y4Vvbk6kkC4LLtJMowkSQxmk4ggVJEG+7c6QpHT8vvW9X7/o7+3ELmiJi2mEzZJiz8cT6TBlanBk70cB5GGIGC1gRDdZ00yADLW1FL6gqhtvNXNG5S9gdSrk4M1qu7JAsmYshzDS4peoMrU/gT7qQdqYGZaYhxZmVbGJAm/CS/HloWyhRUlknQ9KYcExTwS80d3VNOxUZJpITYyspl0LbhArhpZCD9cRWEQuhYkNGMHToQ/2Cs6swJlb39CsllxdXX6IUKh/H5jbnSsPKjgmoaFQ1f8wRLR0UnGE/RcDEjj2jXG1WVTwUs8+zxfcrVO+vSsuOpVKxCfYZiQ0/aPKuxQbQ8lIz+DClxC8u+snlcJ7Yr1z1JPqUH0V+GDXbOwAib931Y4Imaq0NTIXPXY+N5L18GJ37SVWu+hwXff8l72Ds9XuwYIBaXPq6Shm4l+Vl/5QiOlV+uTk6YR9PxKsI9xNJny31ygK1e+nIRC1N97EGkFPI+jCpiHe5PCEy7oWqWSwRrpOvhFzcbTWMbm3ZJAOn1rUKpYIt/lDhW/5RHHteeWFN60qo98YJuoq1nK3uW5AabyspC1BcIEpOhft+SZAShYoLSvnmSfnYADUERP5jJn2h5XtsgCRuhYQqAvwTwn33+YWEKUI72HX5AtfSAZDe8F2DtPPm77afhl0EkthzuCQU0BWApgQIH9+KB0JhopMM7bJrdTRoleM2JAVNMyPF+wdoaz+XJpGoVAQ7WXUkcV7gT3oUZyi/ISIJAVKhgNp+4b4veCFhYVJw4locdSjZCp9cPUhLF9EZ3KKzURepMEtCDPP3VcWFx4UIiZIklIpFNfHpdEafIF2aRmOcrUmjohbT2WUllbmRvgfbythbQO3222fpDJoufaQPncYYuqoGtUEsCJZL6/3PR5b4syeSjZMQG/T2maGANlXT2v8S4AULWaUkCxfLyW8iW4kdka+nEMjxpL2NCwsYNBp+Q61PF43zyDg9Bm9+3NNySn78jMZUUkumqE4Gp7JmFOdP1vc8PpRrzj9+wPinCy8K1PiJ4aYbnTYpCCbDkBSbzhu2QJ1Gd82t8jI8TH51+OzvXoWbnXUOBkNW+0mWFwGcGOUVpU81/n3TOHb5oMt2FgYGjzau0Nif0Ss7Q3XB33hjjQHjHA5E5aOyIQc8CBrLdQSs3j92VG+3nNEjbkbdbBr9zm04ruvw37vh0QKOdeGIkckc80fX3KH/h7PT4BOjgCty8VZ5ux1MoO5Cf5naca2LAsEgehI+drX8o/0Nu+W0m6K/I9gGPd/dfx/EN/wN62AhsBWuAAAAAElFTkSuQmCC
-">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
 </div>
 
-## Overview
 
-The DINOv2 model was proposed in [DINOv2: Learning Robust Visual Features without Supervision](https://arxiv.org/abs/2304.07193) by
-Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski.
-DINOv2 is an upgrade of [DINO](https://arxiv.org/abs/2104.14294), a self-supervised method applied on [Vision Transformers](vit). This method enables all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning.
+# DINOv2
 
-The abstract from the paper is the following:
+[DINOv2](https://huggingface.co/papers/2304.07193) is a vision foundation model that uses [ViT](./vit) as a feature extractor for multiple downstream tasks like image classification and depth estimation. It focuses on stabilizing and accelerating training through techniques like a faster memory-efficient attention, sequence packing, improved stochastic depth, Fully Sharded Data Parallel (FSDP), and model distillation.
 
-*The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could greatly simplify the use of images in any system by producing all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning. This work shows that existing pretraining methods, especially self-supervised methods, can produce such features if trained on enough curated data from diverse sources. We revisit existing approaches and combine different techniques to scale our pretraining in terms of data and model size. Most of the technical contributions aim at accelerating and stabilizing the training at scale. In terms of data, we propose an automatic pipeline to build a dedicated, diverse, and curated image dataset instead of uncurated data, as typically done in the self-supervised literature. In terms of models, we train a ViT model (Dosovitskiy et al., 2020) with 1B parameters and distill it into a series of smaller models that surpass the best available all-purpose features, OpenCLIP (Ilharco et al., 2021) on most of the benchmarks at image and pixel levels.*
+You can find all the original DINOv2 checkpoints under the [Dinov2](https://huggingface.co/collections/facebook/dinov2-6526c98554b3d2576e071ce3) collection.
 
-This model was contributed by [nielsr](https://huggingface.co/nielsr).
-The original code can be found [here](https://github.com/facebookresearch/dinov2).
+> [!TIP]
+> Click on the DINOv2 models in the right sidebar for more examples of how to apply DINOv2 to different vision tasks.
 
-## Usage tips
+The example below demonstrates how to obtain an image embedding with [`Pipeline`] or the [`AutoModel`] class.
 
-The model can be traced using `torch.jit.trace` which leverages JIT compilation to optimize the model making it faster to run. Note this still produces some mis-matched elements and the difference between the original model and the traced model is of the order of 1e-4.
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
-```python
+```py
 import torch
-from transformers import AutoImageProcessor, AutoModel
-from PIL import Image
+from transformers import pipeline
+
+pipe = pipeline(
+    task="image-classification", 
+    model="facebook/dinov2-small-imagenet1k-1-layer",
+    torch_dtype=torch.float16,
+    device=0
+)
+
+pipe("https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/pipeline-cat-chonk.jpeg")
+```
+
+</hfoption>
+<hfoption id="AutoModel">
+
+```py
 import requests
+from transformers import AutoImageProcessor, AutoModelForImageClassification
+from PIL import Image
+
+url = "http://images.cocodataset.org/val2017/000000039769.jpg"
+image = Image.open(requests.get(url, stream=True).raw)
+
+processor = AutoImageProcessor.from_pretrained("facebook/dinov2-small-imagenet1k-1-layer")
+model = AutoModelForImageClassification.from_pretrained(
+    "facebook/dinov2-small-imagenet1k-1-layer", 
+    torch_dtype=torch.float16, 
+    device_map="auto", 
+    attn_implementation="sdpa"
+)
+
+inputs = processor(images=image, return_tensors="pt")
+logits = model(**inputs).logits
+predicted_class_idx = logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
+```
+
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [torchao](../quantization/torchao) to only quantize the weights to int4.
+
+```py
+# pip install torchao
+import requests
+from transformers import TorchAoConfig, AutoImageProcessor, AutoModelForImageClassification
+from torchao.quantization import Int4WeightOnlyConfig
+from PIL import Image
 
 url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
 image = Image.open(requests.get(url, stream=True).raw)
 
-processor = AutoImageProcessor.from_pretrained('facebook/dinov2-base')
-model = AutoModel.from_pretrained('facebook/dinov2-base')
+processor = AutoImageProcessor.from_pretrained('facebook/dinov2-giant-imagenet1k-1-layer')
+
+quant_config = Int4WeightOnlyConfig(group_size=128)
+quantization_config = TorchAoConfig(quant_type=quant_config)
+
+model = AutoModelForImageClassification.from_pretrained(
+    'facebook/dinov2-giant-imagenet1k-1-layer',
+    torch_dtype=torch.bfloat16,
+    device_map="auto",
+    quantization_config=quantization_config
+)
 
 inputs = processor(images=image, return_tensors="pt")
 outputs = model(**inputs)
-last_hidden_states = outputs[0]
-
-# We have to force return_dict=False for tracing
-model.config.return_dict = False
-
-with torch.no_grad():
-    traced_model = torch.jit.trace(model, [inputs.pixel_values])
-    traced_outputs = traced_model(inputs.pixel_values)
-
-print((last_hidden_states - traced_outputs[0]).abs().max())
+logits = outputs.logits
+predicted_class_idx = logits.argmax(-1).item()
+print("Predicted class:", model.config.id2label[predicted_class_idx])
 ```
 
-## Resources
+## Notes
 
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with DINOv2.
+- Use [torch.jit.trace](https://pytorch.org/docs/stable/generated/torch.jit.trace.html) to speedup inference. However, it will produce some mismatched elements. The difference between the original and traced model is 1e-4.
 
-- Demo notebooks for DINOv2 can be found [here](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/DINOv2). 🌎
+    ```py
+    import torch
+    from transformers import AutoImageProcessor, AutoModel
+    from PIL import Image
+    import requests
 
-<PipelineTag pipeline="image-classification"/>
+    url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
+    image = Image.open(requests.get(url, stream=True).raw)
 
-- [`Dinov2ForImageClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/image-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/image_classification.ipynb).
-- See also: [Image classification task guide](../tasks/image_classification)
+    processor = AutoImageProcessor.from_pretrained('facebook/dinov2-base')
+    model = AutoModel.from_pretrained('facebook/dinov2-base')
 
-If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
+    inputs = processor(images=image, return_tensors="pt")
+    outputs = model(**inputs)
+    last_hidden_states = outputs[0]
+
+    # We have to force return_dict=False for tracing
+    model.config.return_dict = False
+
+    with torch.no_grad():
+        traced_model = torch.jit.trace(model, [inputs.pixel_values])
+        traced_outputs = traced_model(inputs.pixel_values)
+
+    print((last_hidden_states - traced_outputs[0]).abs().max())
+    ```
 
 ## Dinov2Config
 

docs/source/en/model_doc/dinov2_with_registers.md

@@ -11,6 +11,7 @@ specific language governing permissions and limitations under the License.
 
 <div class="flex flex-wrap space-x-1">
 <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
 <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
 </div>
 

docs/source/en/model_doc/distilbert.md

@@ -14,199 +14,91 @@ rendered properly in your Markdown viewer.
 
 -->
 
-# DistilBERT
-
-<div class="flex flex-wrap space-x-1">
-<img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
-<img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
-<img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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
-">
-<img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
-<img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="TensorFlow" src="https://img.shields.io/badge/TensorFlow-FF6F00?style=flat&logo=tensorflow&logoColor=white">
+        <img alt="Flax" src="https://img.shields.io/badge/Flax-29a79b.svg?style=flat&logo=data:image/png;base64,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">
+        <img alt="SDPA" src="https://img.shields.io/badge/SDPA-DE3412?style=flat&logo=pytorch&logoColor=white">
+        <img alt="FlashAttention" src="https://img.shields.io/badge/%E2%9A%A1%EF%B8%8E%20FlashAttention-eae0c8?style=flat">
+    </div>
 </div>
 
-## Overview
+# DistilBERT
 
-The DistilBERT model was proposed in the blog post [Smaller, faster, cheaper, lighter: Introducing DistilBERT, a
-distilled version of BERT](https://medium.com/huggingface/distilbert-8cf3380435b5), and the paper [DistilBERT, a
-distilled version of BERT: smaller, faster, cheaper and lighter](https://arxiv.org/abs/1910.01108). DistilBERT is a
-small, fast, cheap and light Transformer model trained by distilling BERT base. It has 40% less parameters than
-*google-bert/bert-base-uncased*, runs 60% faster while preserving over 95% of BERT's performances as measured on the GLUE language
-understanding benchmark.
+[DistilBERT](https://huggingface.co/papers/1910.01108) is pretrained by knowledge distillation to create a smaller model with faster inference and requires less compute to train. Through a triple loss objective during pretraining, language modeling loss, distillation loss, cosine-distance loss, DistilBERT demonstrates similar performance to a larger transformer language model.
 
-The abstract from the paper is the following:
+You can find all the original DistilBERT checkpoints under the [DistilBERT](https://huggingface.co/distilbert) organization.
 
-*As Transfer Learning from large-scale pre-trained models becomes more prevalent in Natural Language Processing (NLP),
-operating these large models in on-the-edge and/or under constrained computational training or inference budgets
-remains challenging. In this work, we propose a method to pre-train a smaller general-purpose language representation
-model, called DistilBERT, which can then be fine-tuned with good performances on a wide range of tasks like its larger
-counterparts. While most prior work investigated the use of distillation for building task-specific models, we leverage
-knowledge distillation during the pretraining phase and show that it is possible to reduce the size of a BERT model by
-40%, while retaining 97% of its language understanding capabilities and being 60% faster. To leverage the inductive
-biases learned by larger models during pretraining, we introduce a triple loss combining language modeling,
-distillation and cosine-distance losses. Our smaller, faster and lighter model is cheaper to pre-train and we
-demonstrate its capabilities for on-device computations in a proof-of-concept experiment and a comparative on-device
-study.*
+> [!TIP]
+> Click on the DistilBERT models in the right sidebar for more examples of how to apply DistilBERT to different language tasks.
 
-This model was contributed by [victorsanh](https://huggingface.co/victorsanh). This model jax version was
-contributed by [kamalkraj](https://huggingface.co/kamalkraj). The original code can be found [here](https://github.com/huggingface/transformers/tree/main/examples/research_projects/distillation).
+The example below demonstrates how to classify text with [`Pipeline`], [`AutoModel`], and from the command line.
 
-## Usage tips
+<hfoptions id="usage">
+
+<hfoption id="Pipeline">
+
+```py
+from transformers import pipeline
+
+classifier = pipeline(
+    task="text-classification",
+    model="distilbert-base-uncased-finetuned-sst-2-english",
+    torch_dtype=torch.float16,
+    device=0
+)
+
+result = classifier("I love using Hugging Face Transformers!")
+print(result)
+# Output: [{'label': 'POSITIVE', 'score': 0.9998}]
+```
+
+</hfoption>
+
+<hfoption id="AutoModel">
+
+```py
+import torch
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+
+tokenizer = AutoTokenizer.from_pretrained(
+    "distilbert/distilbert-base-uncased-finetuned-sst-2-english",
+)
+model = AutoModelForSequenceClassification.from_pretrained(
+    "distilbert/distilbert-base-uncased-finetuned-sst-2-english",
+    torch_dtype=torch.float16,
+    device_map="auto",
+    attn_implementation="sdpa"
+)
+inputs = tokenizer("I love using Hugging Face Transformers!", return_tensors="pt").to("cuda")
+
+with torch.no_grad():
+    outputs = model(**inputs)
+
+predicted_class_id = torch.argmax(outputs.logits, dim=-1).item()
+predicted_label = model.config.id2label[predicted_class_id]
+print(f"Predicted label: {predicted_label}")
+```
+
+</hfoption>
+
+<hfoption id="transformers-cli">
+
+```bash
+echo -e "I love using Hugging Face Transformers!" | transformers-cli run --task text-classification --model distilbert-base-uncased-finetuned-sst-2-english
+```
+
+</hfoption>
+
+</hfoptions>
+
+## Notes
 
 - DistilBERT doesn't have `token_type_ids`, you don't need to indicate which token belongs to which segment. Just
   separate your segments with the separation token `tokenizer.sep_token` (or `[SEP]`).
 - DistilBERT doesn't have options to select the input positions (`position_ids` input). This could be added if
   necessary though, just let us know if you need this option.
-- Same as BERT but smaller. Trained by distillation of the pretrained BERT model, meaning it’s been trained to predict the same probabilities as the larger model. The actual objective is a combination of:
-
-    * finding the same probabilities as the teacher model
-    * predicting the masked tokens correctly (but no next-sentence objective)
-    * a cosine similarity between the hidden states of the student and the teacher model
-
-### Using Scaled Dot Product Attention (SDPA)
-
-PyTorch includes a native scaled dot-product attention (SDPA) operator as part of `torch.nn.functional`. This function 
-encompasses several implementations that can be applied depending on the inputs and the hardware in use. See the 
-[official documentation](https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html) 
-or the [GPU Inference](https://huggingface.co/docs/transformers/main/en/perf_infer_gpu_one#pytorch-scaled-dot-product-attention)
-page for more information.
-
-SDPA is used by default for `torch>=2.1.1` when an implementation is available, but you may also set 
-`attn_implementation="sdpa"` in `from_pretrained()` to explicitly request SDPA to be used.
-
-```
-from transformers import DistilBertModel
-model = DistilBertModel.from_pretrained("distilbert-base-uncased", torch_dtype=torch.float16, attn_implementation="sdpa")
-```
-
-For the best speedups, we recommend loading the model in half-precision (e.g. `torch.float16` or `torch.bfloat16`).
-
-On a local benchmark (NVIDIA GeForce RTX 2060-8GB, PyTorch 2.3.1, OS Ubuntu 20.04) with `float16` and the `distilbert-base-uncased` model with
-a MaskedLM head, we saw the following speedups during training and inference.
-
-#### Training
-
-| num_training_steps | batch_size | seq_len | is cuda | Time per batch (eager - s) | Time per batch (sdpa - s) | Speedup (%) | Eager peak mem (MB) | sdpa peak mem (MB) | Mem saving (%) |
-|--------------------|------------|---------|---------|----------------------------|---------------------------|-------------|---------------------|--------------------|----------------|
-| 100                | 1          | 128     | False   | 0.010                      | 0.008                     | 28.870      | 397.038             | 399.629            | -0.649         |
-| 100                | 1          | 256     | False   | 0.011                      | 0.009                     | 20.681      | 412.505             | 412.606            | -0.025         |
-| 100                | 2          | 128     | False   | 0.011                      | 0.009                     | 23.741      | 412.213             | 412.606            | -0.095         |
-| 100                | 2          | 256     | False   | 0.015                      | 0.013                     | 16.502      | 427.491             | 425.787            | 0.400          |
-| 100                | 4          | 128     | False   | 0.015                      | 0.013                     | 13.828      | 427.491             | 425.787            | 0.400          |
-| 100                | 4          | 256     | False   | 0.025                      | 0.022                     | 12.882      | 594.156             | 502.745            | 18.182         |
-| 100                | 8          | 128     | False   | 0.023                      | 0.022                     | 8.010       | 545.922             | 502.745            | 8.588          |
-| 100                | 8          | 256     | False   | 0.046                      | 0.041                     | 12.763      | 983.450             | 798.480            | 23.165         |
-
-#### Inference
-
-| num_batches | batch_size | seq_len | is cuda | is half | use mask | Per token latency eager (ms) | Per token latency SDPA (ms) | Speedup (%) | Mem eager (MB) | Mem BT (MB) | Mem saved (%) |
-|-------------|------------|---------|---------|---------|----------|-----------------------------|-----------------------------|-------------|----------------|--------------|---------------|
-| 50          | 2          | 64      | True    | True    | True     | 0.032                       | 0.025                       | 28.192      | 154.532        | 155.531      | -0.642        |
-| 50          | 2          | 128     | True    | True    | True     | 0.033                       | 0.025                       | 32.636      | 157.286        | 157.482      | -0.125        |
-| 50          | 4          | 64      | True    | True    | True     | 0.032                       | 0.026                       | 24.783      | 157.023        | 157.449      | -0.271        |
-| 50          | 4          | 128     | True    | True    | True     | 0.034                       | 0.028                       | 19.299      | 162.794        | 162.269      | 0.323         |
-| 50          | 8          | 64      | True    | True    | True     | 0.035                       | 0.028                       | 25.105      | 160.958        | 162.204      | -0.768        |
-| 50          | 8          | 128     | True    | True    | True     | 0.052                       | 0.046                       | 12.375      | 173.155        | 171.844      | 0.763         |
-| 50          | 16         | 64      | True    | True    | True     | 0.051                       | 0.045                       | 12.882      | 172.106        | 171.713      | 0.229         |
-| 50          | 16         | 128     | True    | True    | True     | 0.096                       | 0.081                       | 18.524      | 191.257        | 191.517      | -0.136        |
-
-
-## Resources
-
-A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with DistilBERT. If you're interested in submitting a resource to be included here, please feel free to open a Pull Request and we'll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
-
-<PipelineTag pipeline="text-classification"/>
-
-- A blog post on [Getting Started with Sentiment Analysis using Python](https://huggingface.co/blog/sentiment-analysis-python) with DistilBERT.
-- A blog post on how to [train DistilBERT with Blurr for sequence classification](https://huggingface.co/blog/fastai).
-- A blog post on how to use [Ray to tune DistilBERT hyperparameters](https://huggingface.co/blog/ray-tune).
-- A blog post on how to [train DistilBERT with Hugging Face and Amazon SageMaker](https://huggingface.co/blog/the-partnership-amazon-sagemaker-and-hugging-face).
-- A notebook on how to [finetune DistilBERT for multi-label classification](https://colab.research.google.com/github/DhavalTaunk08/Transformers_scripts/blob/master/Transformers_multilabel_distilbert.ipynb). 🌎
-- A notebook on how to [finetune DistilBERT for multiclass classification with PyTorch](https://colab.research.google.com/github/abhimishra91/transformers-tutorials/blob/master/transformers_multiclass_classification.ipynb). 🌎
-- A notebook on how to [finetune DistilBERT for text classification in TensorFlow](https://colab.research.google.com/github/peterbayerle/huggingface_notebook/blob/main/distilbert_tf.ipynb). 🌎
-- [`DistilBertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification.ipynb).
-- [`TFDistilBertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification-tf.ipynb).
-- [`FlaxDistilBertForSequenceClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/text-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/text_classification_flax.ipynb).
-- [Text classification task guide](../tasks/sequence_classification)
-
-
-<PipelineTag pipeline="token-classification"/>
-
-- [`DistilBertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification.ipynb).
-- [`TFDistilBertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/token-classification) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/token_classification-tf.ipynb).
-- [`FlaxDistilBertForTokenClassification`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/token-classification).
-- [Token classification](https://huggingface.co/course/chapter7/2?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Token classification task guide](../tasks/token_classification)
-
-
-<PipelineTag pipeline="fill-mask"/>
-
-- [`DistilBertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/language-modeling#robertabertdistilbert-and-masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling.ipynb).
-- [`TFDistilBertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/language-modeling#run_mlmpy) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/language_modeling-tf.ipynb).
-- [`FlaxDistilBertForMaskedLM`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/language-modeling#masked-language-modeling) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/masked_language_modeling_flax.ipynb).
-- [Masked language modeling](https://huggingface.co/course/chapter7/3?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Masked language modeling task guide](../tasks/masked_language_modeling)
-
-<PipelineTag pipeline="question-answering"/>
-
-- [`DistilBertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering.ipynb).
-- [`TFDistilBertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/question-answering) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/question_answering-tf.ipynb).
-- [`FlaxDistilBertForQuestionAnswering`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/flax/question-answering).
-- [Question answering](https://huggingface.co/course/chapter7/7?fw=pt) chapter of the 🤗 Hugging Face Course.
-- [Question answering task guide](../tasks/question_answering)
-
-**Multiple choice**
-- [`DistilBertForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/pytorch/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice.ipynb).
-- [`TFDistilBertForMultipleChoice`] is supported by this [example script](https://github.com/huggingface/transformers/tree/main/examples/tensorflow/multiple-choice) and [notebook](https://colab.research.google.com/github/huggingface/notebooks/blob/main/examples/multiple_choice-tf.ipynb).
-- [Multiple choice task guide](../tasks/multiple_choice)
-
-⚗️ Optimization
-
-- A blog post on how to [quantize DistilBERT with 🤗 Optimum and Intel](https://huggingface.co/blog/intel).
-- A blog post on how [Optimizing Transformers for GPUs with 🤗 Optimum](https://www.philschmid.de/optimizing-transformers-with-optimum-gpu).
-- A blog post on [Optimizing Transformers with Hugging Face Optimum](https://www.philschmid.de/optimizing-transformers-with-optimum).
-
-⚡️ Inference
-
-- A blog post on how to [Accelerate BERT inference with Hugging Face Transformers and AWS Inferentia](https://huggingface.co/blog/bert-inferentia-sagemaker) with DistilBERT.
-- A blog post on [Serverless Inference with Hugging Face's Transformers, DistilBERT and Amazon SageMaker](https://www.philschmid.de/sagemaker-serverless-huggingface-distilbert).
-
-🚀 Deploy
-
-- A blog post on how to [deploy DistilBERT on Google Cloud](https://huggingface.co/blog/how-to-deploy-a-pipeline-to-google-clouds).
-- A blog post on how to [deploy DistilBERT with Amazon SageMaker](https://huggingface.co/blog/deploy-hugging-face-models-easily-with-amazon-sagemaker).
-- A blog post on how to [Deploy BERT with Hugging Face Transformers, Amazon SageMaker and Terraform module](https://www.philschmid.de/terraform-huggingface-amazon-sagemaker).
-
-
-## Combining DistilBERT and Flash Attention 2
-
-First, make sure to install the latest version of Flash Attention 2 to include the sliding window attention feature.
-
-```bash
-pip install -U flash-attn --no-build-isolation
-```
-
-Make also sure that you have a hardware that is compatible with Flash-Attention 2. Read more about it in the official documentation of flash-attn repository. Make also sure to load your model in half-precision (e.g. `torch.float16`)
-
-To load and run a model using Flash Attention 2, refer to the snippet below:
-
-```python
->>> import torch
->>> from transformers import AutoTokenizer, AutoModel
-
->>> device = "cuda" # the device to load the model onto
-
->>> tokenizer = AutoTokenizer.from_pretrained('distilbert/distilbert-base-uncased')
->>> model = AutoModel.from_pretrained("distilbert/distilbert-base-uncased", torch_dtype=torch.float16, attn_implementation="flash_attention_2")
-
->>> text = "Replace me by any text you'd like."
-
->>> encoded_input = tokenizer(text, return_tensors='pt').to(device)
->>> model.to(device)
-
->>> output = model(**encoded_input)
-```
-
 
 ## DistilBertConfig
 

docs/source/en/model_doc/donut.md

@@ -13,180 +13,191 @@ rendered properly in your Markdown viewer.
 
 specific language governing permissions and limitations under the License. -->
 
+<div style="float: right;">
+    <div class="flex flex-wrap space-x-1">
+        <img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-DE3412?style=flat&logo=pytorch&logoColor=white">
+    </div>
+</div>
+
 # Donut
 
-## Overview
+[Donut (Document Understanding Transformer)](https://huggingface.co/papers2111.15664) is a visual document understanding model that doesn't require an Optical Character Recognition (OCR) engine. Unlike traditional approaches that extract text using OCR before processing, Donut employs an end-to-end Transformer-based architecture to directly analyze document images. This eliminates OCR-related inefficiencies making it more accurate and adaptable to diverse languages and formats. 
 
-The Donut model was proposed in [OCR-free Document Understanding Transformer](https://arxiv.org/abs/2111.15664) by
-Geewook Kim, Teakgyu Hong, Moonbin Yim, Jeongyeon Nam, Jinyoung Park, Jinyeong Yim, Wonseok Hwang, Sangdoo Yun, Dongyoon Han, Seunghyun Park.
-Donut consists of an image Transformer encoder and an autoregressive text Transformer decoder to perform document understanding
-tasks such as document image classification, form understanding and visual question answering.
+Donut features vision encoder ([Swin](./swin)) and a text decoder ([BART](./bart)). Swin converts document images into embeddings and BART processes them into meaningful text sequences.
 
-The abstract from the paper is the following:
+You can find all the original Donut checkpoints under the [Naver Clova Information Extraction](https://huggingface.co/naver-clova-ix) organization.
 
-*Understanding document images (e.g., invoices) is a core but challenging task since it requires complex functions such as reading text and a holistic understanding of the document. Current Visual Document Understanding (VDU) methods outsource the task of reading text to off-the-shelf Optical Character Recognition (OCR) engines and focus on the understanding task with the OCR outputs. Although such OCR-based approaches have shown promising performance, they suffer from 1) high computational costs for using OCR; 2) inflexibility of OCR models on languages or types of document; 3) OCR error propagation to the subsequent process. To address these issues, in this paper, we introduce a novel OCR-free VDU model named Donut, which stands for Document understanding transformer. As the first step in OCR-free VDU research, we propose a simple architecture (i.e., Transformer) with a pre-training objective (i.e., cross-entropy loss). Donut is conceptually simple yet effective. Through extensive experiments and analyses, we show a simple OCR-free VDU model, Donut, achieves state-of-the-art performances on various VDU tasks in terms of both speed and accuracy. In addition, we offer a synthetic data generator that helps the model pre-training to be flexible in various languages and domains.*
+> [!TIP]
+> Click on the Donut models in the right sidebar for more examples of how to apply Donut to different language and vision tasks.
 
-<img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/model_doc/donut_architecture.jpg"
-alt="drawing" width="600"/>
+The examples below demonstrate how to perform document understanding tasks using Donut with [`Pipeline`] and [`AutoModel`]
 
-<small> Donut high-level overview. Taken from the <a href="https://arxiv.org/abs/2111.15664">original paper</a>. </small>
-
-This model was contributed by [nielsr](https://huggingface.co/nielsr). The original code can be found
-[here](https://github.com/clovaai/donut).
-
-## Usage tips
-
-- The quickest way to get started with Donut is by checking the [tutorial
-  notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Donut), which show how to use the model
-  at inference time as well as fine-tuning on custom data.
-- Donut is always used within the [VisionEncoderDecoder](vision-encoder-decoder) framework.
-
-## Inference examples
-
-Donut's [`VisionEncoderDecoder`] model accepts images as input and makes use of
-[`~generation.GenerationMixin.generate`] to autoregressively generate text given the input image.
-
-The [`DonutImageProcessor`] class is responsible for preprocessing the input image and
-[`XLMRobertaTokenizer`/`XLMRobertaTokenizerFast`] decodes the generated target tokens to the target string. The
-[`DonutProcessor`] wraps [`DonutImageProcessor`] and [`XLMRobertaTokenizer`/`XLMRobertaTokenizerFast`]
-into a single instance to both extract the input features and decode the predicted token ids.
-
-- Step-by-step Document Image Classification
+<hfoptions id="usage">
+<hfoption id="Pipeline">
 
 ```py
->>> import re
+# pip install datasets
+import torch
+from transformers import pipeline
+from PIL import Image
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+pipeline = pipeline(
+    task="document-question-answering",
+    model="naver-clova-ix/donut-base-finetuned-docvqa",
+    device=0,
+    torch_dtype=torch.float16
+)
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
-
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[1]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_rvlcdip>"
->>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'class': 'advertisement'}
+pipeline(image=image, question="What time is the coffee break?")
 ```
 
-- Step-by-step Document Parsing
+</hfoption>
+<hfoption id="AutoModel">
 
 ```py
->>> import re
+# pip install datasets
+import torch
+from datasets import load_dataset
+from transformers import AutoProcessor, AutoModelForVision2Seq
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+model = AutoModelForVision2Seq.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
+question = "What time is the coffee break?"
+task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
+inputs = processor(image, task_prompt, return_tensors="pt")
 
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[2]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_cord-v2>"
->>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'menu': {'nm': 'CINNAMON SUGAR', 'unitprice': '17,000', 'cnt': '1 x', 'price': '17,000'}, 'sub_total': {'subtotal_price': '17,000'}, 'total': {'total_price': '17,000', 'cashprice': '20,000', 'changeprice': '3,000'}}
+outputs = model.generate(
+    input_ids=inputs.input_ids,
+    pixel_values=inputs.pixel_values,
+    max_length=512
+)
+answer = processor.decode(outputs[0], skip_special_tokens=True)
+print(answer)
 ```
 
-- Step-by-step Document Visual Question Answering (DocVQA)
+</hfoption>
+</hfoptions>
+
+Quantization reduces the memory burden of large models by representing the weights in a lower precision. Refer to the [Quantization](../quantization/overview) overview for more available quantization backends.
+
+The example below uses [torchao](../quantization/torchao) to only quantize the weights to int4.
 
 ```py
->>> import re
+# pip install datasets torchao
+import torch
+from datasets import load_dataset
+from transformers import TorchAoConfig, AutoProcessor, AutoModelForVision2Seq
 
->>> from transformers import DonutProcessor, VisionEncoderDecoderModel
->>> from datasets import load_dataset
->>> import torch
+quantization_config = TorchAoConfig("int4_weight_only", group_size=128)
+processor = AutoProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+model = AutoModelForVision2Seq.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa", quantization_config=quantization_config)
 
->>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
->>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-docvqa")
+dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+image = dataset[0]["image"]
+question = "What time is the coffee break?"
+task_prompt = f"<s_docvqa><s_question>{question}</s_question><s_answer>"
+inputs = processor(image, task_prompt, return_tensors="pt")
 
->>> device = "cuda" if torch.cuda.is_available() else "cpu"
->>> model.to(device)  # doctest: +IGNORE_RESULT
-
->>> # load document image from the DocVQA dataset
->>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
->>> image = dataset[0]["image"]
-
->>> # prepare decoder inputs
->>> task_prompt = "<s_docvqa><s_question>{user_input}</s_question><s_answer>"
->>> question = "When is the coffee break?"
->>> prompt = task_prompt.replace("{user_input}", question)
->>> decoder_input_ids = processor.tokenizer(prompt, add_special_tokens=False, return_tensors="pt").input_ids
-
->>> pixel_values = processor(image, return_tensors="pt").pixel_values
-
->>> outputs = model.generate(
-...     pixel_values.to(device),
-...     decoder_input_ids=decoder_input_ids.to(device),
-...     max_length=model.decoder.config.max_position_embeddings,
-...     pad_token_id=processor.tokenizer.pad_token_id,
-...     eos_token_id=processor.tokenizer.eos_token_id,
-...     use_cache=True,
-...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
-...     return_dict_in_generate=True,
-... )
-
->>> sequence = processor.batch_decode(outputs.sequences)[0]
->>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
->>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
->>> print(processor.token2json(sequence))
-{'question': 'When is the coffee break?', 'answer': '11-14 to 11:39 a.m.'}
+outputs = model.generate(
+    input_ids=inputs.input_ids,
+    pixel_values=inputs.pixel_values,
+    max_length=512
+)
+answer = processor.decode(outputs[0], skip_special_tokens=True)
+print(answer)
 ```
 
-See the [model hub](https://huggingface.co/models?filter=donut) to look for Donut checkpoints.
+## Notes
 
-## Training
+- Use Donut for document image classification as shown below.
 
-We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-Tutorials/tree/master/Donut).
+    ```py
+    >>> import re
+    >>> from transformers import DonutProcessor, VisionEncoderDecoderModel
+    >>> from datasets import load_dataset
+    >>> import torch
+
+    >>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
+    >>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-rvlcdip")
+
+    >>> device = "cuda" if torch.cuda.is_available() else "cpu"
+    >>> model.to(device)  # doctest: +IGNORE_RESULT
+
+    >>> # load document image
+    >>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+    >>> image = dataset[1]["image"]
+
+    >>> # prepare decoder inputs
+    >>> task_prompt = "<s_rvlcdip>"
+    >>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
+
+    >>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    >>> outputs = model.generate(
+    ...     pixel_values.to(device),
+    ...     decoder_input_ids=decoder_input_ids.to(device),
+    ...     max_length=model.decoder.config.max_position_embeddings,
+    ...     pad_token_id=processor.tokenizer.pad_token_id,
+    ...     eos_token_id=processor.tokenizer.eos_token_id,
+    ...     use_cache=True,
+    ...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+    ...     return_dict_in_generate=True,
+    ... )
+
+    >>> sequence = processor.batch_decode(outputs.sequences)[0]
+    >>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
+    >>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
+    >>> print(processor.token2json(sequence))
+    {'class': 'advertisement'}
+    ```
+
+- Use Donut for document parsing as shown below.
+
+    ```py
+    >>> import re
+    >>> from transformers import DonutProcessor, VisionEncoderDecoderModel
+    >>> from datasets import load_dataset
+    >>> import torch
+
+    >>> processor = DonutProcessor.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+    >>> model = VisionEncoderDecoderModel.from_pretrained("naver-clova-ix/donut-base-finetuned-cord-v2")
+
+    >>> device = "cuda" if torch.cuda.is_available() else "cpu"
+    >>> model.to(device)  # doctest: +IGNORE_RESULT
+
+    >>> # load document image
+    >>> dataset = load_dataset("hf-internal-testing/example-documents", split="test")
+    >>> image = dataset[2]["image"]
+
+    >>> # prepare decoder inputs
+    >>> task_prompt = "<s_cord-v2>"
+    >>> decoder_input_ids = processor.tokenizer(task_prompt, add_special_tokens=False, return_tensors="pt").input_ids
+
+    >>> pixel_values = processor(image, return_tensors="pt").pixel_values
+
+    >>> outputs = model.generate(
+    ...     pixel_values.to(device),
+    ...     decoder_input_ids=decoder_input_ids.to(device),
+    ...     max_length=model.decoder.config.max_position_embeddings,
+    ...     pad_token_id=processor.tokenizer.pad_token_id,
+    ...     eos_token_id=processor.tokenizer.eos_token_id,
+    ...     use_cache=True,
+    ...     bad_words_ids=[[processor.tokenizer.unk_token_id]],
+    ...     return_dict_in_generate=True,
+    ... )
+
+    >>> sequence = processor.batch_decode(outputs.sequences)[0]
+    >>> sequence = sequence.replace(processor.tokenizer.eos_token, "").replace(processor.tokenizer.pad_token, "")
+    >>> sequence = re.sub(r"<.*?>", "", sequence, count=1).strip()  # remove first task start token
+    >>> print(processor.token2json(sequence))
+    {'menu': {'nm': 'CINNAMON SUGAR', 'unitprice': '17,000', 'cnt': '1 x', 'price': '17,000'}, 'sub_total': {'subtotal_price': '17,000'}, 'total': 
+    {'total_price': '17,000', 'cashprice': '20,000', 'changeprice': '3,000'}}
+    ```
 
 ## DonutSwinConfig
 
@@ -197,6 +208,11 @@ We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-
 [[autodoc]] DonutImageProcessor
     - preprocess
 
+## DonutImageProcessorFast
+
+[[autodoc]] DonutImageProcessorFast
+    - preprocess
+
 ## DonutFeatureExtractor
 
 [[autodoc]] DonutFeatureExtractor
@@ -215,3 +231,8 @@ We refer to the [tutorial notebooks](https://github.com/NielsRogge/Transformers-
 
 [[autodoc]] DonutSwinModel
     - forward
+
+## DonutSwinForImageClassification
+
+[[autodoc]] transformers.DonutSwinForImageClassification
+    - forward