b9bd8c45a1
[CI] Build translated docs ( #41632 )
...
fix
2025-10-16 14:01:33 +02:00
baecdb8a97
[Ernie 4.5 Moe] Fix Moe and offloading ( #41385 )
...
fix
2025-10-16 13:59:01 +02:00
44539827d5
[Executorch] Simplify for encoder models ( #41627 )
...
* Trigger Build
* revert extra treatment for executorch as we default to no vmapping now
2025-10-16 13:57:52 +02:00
143acfe2ce
fix check inputs for text2text pipeline ( #41556 )
...
fix check inputs
Signed-off-by: jiqing-feng <jiqing.feng@intel.com >
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com >
2025-10-16 11:42:41 +00:00
67fae90519
Fix FP-Quant quantization fallback CPU dispatch. ( #41619 )
...
* fp_quant fix
* Update quantizer_fp_quant.py
2025-10-16 11:41:01 +00:00
af2a66ced9
Migrate transformers cli to Typer ( #41487 )
...
* Add typer-slim as explicit dependency
* Migrate CLI to Typer
* code quality
* bump release candidate
* adapt test_cli.py
* Remove ./commands + adapt tests
* fix quality
* consistency
* doctested
* do not serve model in chat
* style
* will it fix them?
* fix test
* capitalize classes
* Rebase
* Rebase
* tests + fixup
tests + fixup
* csutom error message
* fix ?
* should be good
* fix caplog globally
* inner caplog
* last attempt
* Retry
* Let's try with capsys disabled
---------
Co-authored-by: Lysandre <hi@lysand.re >
2025-10-16 13:29:42 +02:00
a59124e27e
Add missing dates to docs ( #41576 )
...
add dates
2025-10-16 09:32:28 +00:00
81f97b17d2
Remove randomly added script ( #41650 )
...
remove
2025-10-16 11:23:53 +02:00
c0a5cf19ad
Fix tokenization test ( #41649 )
...
fix
2025-10-16 11:14:20 +02:00
3ef6f2c415
Allow passing tp_plan in from_pretrained directly ( #41435 )
...
* start
* allow passing it
* fix plans
* fix
* fix
* style
* style
* fix
* add_test
* oupsi indent
* fix
* fix
* fix for CI without accelerator
* fix import
2025-10-16 11:12:07 +02:00
59efd86da2
Add aux loss for GLM-4.5V ( #41564 )
...
* add aux
* update
* update config to text_config
* use qwen data class to avoid repeat again
* format
* update
* use 1e-4
* update
* update for remove init
* Apply style fixes
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Raushan Turganbay <raushan@huggingface.co >
2025-10-16 09:04:21 +00:00
7b7d17f9bf
🚨 [v5] Toggle the serialization format in processors ( #41474 )
...
* toggle the serialization
* prob this fixes it
* fix tests
* typo
* delete legacy save entirely
* remove extra nesting in if
* revert test and serialzie a public attr instead of private
2025-10-16 10:19:22 +02:00
e20df45bf6
Add Backbone API fine-tuning tutorial ( #41590 )
...
---------
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com >
2025-10-15 18:42:32 +02:00
19df66dcba
Update executorch.md ( #41582 )
...
* Update executorch.md
* Update executorch.md
* Update executorch.md
* Apply suggestions from code review
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com >
---------
Co-authored-by: Marc Sun <57196510+SunMarc@users.noreply.github.com >
Co-authored-by: Steven Liu <59462357+stevhliu@users.noreply.github.com >
2025-10-15 09:01:46 -07:00
9f71e3a604
[docs] Duplicate entry ( #41591 )
...
fix
2025-10-15 17:02:36 +02:00
bc9900562d
Fix quantization base class ( #41613 )
...
* fix
* fix
---------
Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com >
2025-10-15 16:58:17 +02:00
72fd67929b
Remove deprecated code ( #41616 )
...
remove
Co-authored-by: Mohamed Mekkouri <93391238+MekkCyber@users.noreply.github.com >
2025-10-15 16:57:52 +02:00
da382917aa
Remove the head masking block in some vision models ( #41620 )
...
* old
* new
---------
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com >
2025-10-15 15:51:01 +02:00
313afcc468
[chat template] update when "push_to_hub" ( #39815 )
...
* update templates push to hub
* rvert jinja suffix and move it to processor file
2025-10-15 13:49:59 +00:00
7bba4d1202
Fix video processing channel format ( #41603 )
...
fix
2025-10-15 15:48:01 +02:00
ab92534377
enable sdpa enable gqa logic for Ascend NPU ( #41601 )
...
* enable gqa logic for Ascend NPU
* remove redundant comments
* fix comments about Ascend NPU
---------
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com >
2025-10-15 13:45:28 +00:00
56a727dde5
Add fast path for bidirectional mask creation to fix regression ( #41586 )
...
* fixed performance regression
* also fixed the older_torch function
* Update src/transformers/masking_utils.py
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com >
* fix
* more general
* fix slicing
* fix data dependent
---------
Co-authored-by: Anton Vlasjuk <73884904+vasqu@users.noreply.github.com >
Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com >
Co-authored-by: Cyril Vallez <cyril.vallez@huggingface.co >
2025-10-15 15:30:39 +02:00
dc6fdeb705
Update a dataset reop link ( #41618 )
...
fix
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com >
2025-10-15 14:41:38 +02:00
3953b65440
Reinstate early CUDA init fix ( #41617 )
...
* Reinstate early CUDA init fix
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com >
* Delay import further
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com >
---------
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com >
2025-10-15 14:41:10 +02:00
96d245a83d
torch 2.9 don't ❤️ torchcodec 💔 ( #41610 )
...
pin
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com >
2025-10-15 14:34:00 +02:00
bb0c3af995
More markdown file fixes ( #41599 )
...
* Format markdown files
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
* Format markdown files
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
* Format markdown files
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
---------
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
2025-10-15 12:29:27 +00:00
70e871959c
Fix trainer simple tests ( #41449 )
...
* fix
* fix ray
* train to tune
* breaking changes wrt generation config
* Fix !
* fix
* fix
* fix deepspeed !
* fix
* fix
* fix
* improve logic
* revert and fix
* revert comment
* oups
* revert change
* fix
* style
* typo in comment
---------
Co-authored-by: Cyril Vallez <cyril.vallez@gmail.com >
2025-10-15 14:09:00 +02:00
c4210796e0
Import expand_device_map instead of redefining it ( #41608 )
...
remove it
2025-10-15 14:00:09 +02:00
fcd1ccdb78
[Docs] Fix changed references ( #41614 )
...
* fix
* fix
* other ln
2025-10-15 13:59:13 +02:00
2b2c20f315
Update issue template ( #41573 )
...
* update
* fix
2025-10-15 13:54:37 +02:00
e2122c4bcb
remove ray_scope and check_quantized_param ( #41587 )
...
remove
2025-10-15 13:10:35 +02:00
e89cef6625
fix some case failures lead by "torch.compile recompiled part of th… ( #41558 )
...
* fix some case failures lead by "`torch.compile` recompiled part of the forward pass" in xpu
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com >
* update comment
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com >
---------
Signed-off-by: Wang, Yi A <yi.a.wang@intel.com >
2025-10-15 10:45:29 +00:00
26b7f66850
Add logits_to_keep to many older CausalLM models ( #41335 )
...
* Add logits_to_keep to CausalLM models
* Skip failing test for git model
* Remove unused return_dict from kosmos2 signature
* Revert BlipForQuestionAnswering
2025-10-15 11:56:01 +02:00
5db730786d
[device_map] Accelerate loading by computing device_map much faster ( #41548 )
...
* start
* add the important fix
* continue
* big cleanup
* type hints
* add method
* fix typehints
* typehints
* fix
* oupsi
* remove space
* improve function
* CI
2025-10-15 11:18:57 +02:00
13a35a5057
Enable non-streaming mode in transformers serve ( #41446 )
...
* Enable non-streaming in transformers serve
Remove typos
Remove typos
Remove typos
* Fix tests
* Arthur review
2025-10-15 09:37:26 +02:00
94df0e6560
Benchmark overhaul ( #41408 )
...
* Big refactor, still classes to move around and script to re-complexify
* Move to streamer, isolate benches, propagate num tokens
* Some refacto
* Added compile mode to name
* Re-order
* Move to dt_tokens
* Better format
* Fix and disable use_cache by default
* Fixed compile and SDPA backend default
* Refactor results format
* Added default compile mode
* Always use cache
* Fixed cache and added flex
* Plan for missing modules
* Experiments: no cg and shuffle
* Disable compile for FA
* Remove wall time, add sweep mode, get git commit
* Review compliance, start
* Apply suggestions from code review
Co-authored-by: Luc Georges <McPatate@users.noreply.github.com >
* Update benchmark_v2/framework/benchmark_runner.py
Co-authored-by: Luc Georges <McPatate@users.noreply.github.com >
* Disable workflow
* Pretty print
* Added some pretty names to have pretty logs
* Review n2 compliance (end?)
* Style and end of PR
---------
Co-authored-by: Luc Georges <McPatate@users.noreply.github.com >
2025-10-14 21:41:43 +02:00
9e4199ede3
Gemma3 fixes ( #41572 )
...
* Multiple device error fix
* FA2 equivalence fix
* Move the train fwd in cfg test
* Style
* Added comment
* Made the comment more clear
2025-10-14 18:33:27 +02:00
4c8d293599
Fix typsetting and content of llm_tutorial_optimization.md ( #41172 )
...
* Fix typsetting of llm_tutorial_optimization
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
* Fix errors
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
---------
Signed-off-by: Yuanyuan Chen <cyyever@outlook.com >
2025-10-14 08:40:26 -07:00
a99b1be3c7
Revert some breaking changes bnb ( #41581 )
...
fix
2025-10-14 16:28:16 +02:00
82cae9eb52
Add __iter__ to DynamicCache ( #41569 )
...
* Add __iter__ to DynamicCache
* Fix tests that use ddp init
2025-10-14 16:16:32 +02:00
4fad35ee4a
[VisionEncoderDecoderModel] Update loss function ( #40863 )
...
Update loss function
2025-10-14 16:03:00 +02:00
ae6f6cc3e0
Revert "add rmsnorm kernels support for Intel XPU" ( #41579 )
...
Revert "add rmsnorm kernels support for Intel XPU (#41563 )"
This reverts commit fd787c5f6d667d3e00def70f588972af4437f631.
2025-10-14 15:49:33 +02:00
fd787c5f6d
add rmsnorm kernels support for Intel XPU ( #41563 )
...
Signed-off-by: Liu, Kaixuan <kaixuan.liu@intel.com >
2025-10-14 13:26:09 +00:00
4e4f2af586
Add conditional checks to _check_and_adjust_attn_implementation() ( #41542 )
2025-10-14 13:00:07 +00:00
3648fde486
Add DINOv3Backbone for ConvNext variant ( #40651 )
...
---------
Co-authored-by: Pavel Iakubovskii <qubvel@gmail.com >
2025-10-14 14:57:04 +02:00
abf5b57a68
delete some tokenizer tests using pickle ( #41514 )
...
* hate pickle
* hate pickle
* hate pickle
---------
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com >
2025-10-14 14:50:51 +02:00
8fe4db5399
[kernels] rm mra kernels ( #41507 )
...
* fix modeling
* remove kernel
* fix style
2025-10-14 13:34:04 +02:00
c620c38bb0
[Qwen3VLMoe] Fixed: Expected self.dtype to be equal to src.dtype - routing_weights casting ( #41420 )
...
* Fixed Expected self.dtype to be equal to src.dtype on eval
* Fixed Expected self.dtype to be equal to src.dtype on eval
* Fixed Expected self.dtype to be equal to src.dtype on eval
* generated modeling_qwen3_vl_moe.py file
* Fixed Ernie_4_5_MoE router casting
* Fixed routing_weights dtype casting (ernie4_5_moe, hunyuan_v1_moe, qwen2_moe, qwen3_moe, qwen3_next,qwen3_omni_moe)
* rollback hunyuan_v1_moe changes
---------
Co-authored-by: Daniel Oliveira <daniel-oliveira-11@hotmail.com >
Co-authored-by: Daniel Oliveira <36623265+daniel3303@users.noreply.github.com >
2025-10-14 13:14:49 +02:00
0798797ec9
Fix an import error with PreTrainModel ( #41571 )
2025-10-14 13:13:37 +02:00
0566b6f5bd
Patch MistralCommonTokenizer ( #41439 )
...
* Fix token_to_id and add add_generation_prompt
* Fix spm download
* Refactor spm
* Try another possibly non-gated spm
* Improve get_vocab
* lint
* Improve get_vocab
* Add warn to piece_to_id
* Improve from_pretrained raise and revert model spm
* Revert fast
2025-10-14 11:13:19 +00:00
