[CI/Build] Consolidate model loader tests and requirements (#25765)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

.buildkite/test-pipeline.yaml

@@ -465,29 +465,18 @@ steps:
   commands:
     - pytest -v -s kernels/mamba
 
-- label: Tensorizer Test # 14min
-  timeout_in_minutes: 25
-  mirror_hardwares: [amdexperimental]
-  source_file_dependencies:
-  - vllm/model_executor/model_loader
-  - tests/tensorizer_loader
-  - tests/entrypoints/openai/test_tensorizer_entrypoint.py
-  commands:
-    - apt-get update && apt-get install -y curl libsodium23
-    - export VLLM_WORKER_MULTIPROC_METHOD=spawn
-    - pytest -v -s tensorizer_loader
-    - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py
-
-- label: Model Executor Test # 7min
-  timeout_in_minutes: 20
+- label: Model Executor Test # ???
+  timeout_in_minutes: 60
   mirror_hardwares: [amdexperimental]
   source_file_dependencies:
   - vllm/model_executor
   - tests/model_executor
+  - tests/entrypoints/openai/test_tensorizer_entrypoint.py
   commands:
     - apt-get update && apt-get install -y curl libsodium23
     - export VLLM_WORKER_MULTIPROC_METHOD=spawn
     - pytest -v -s model_executor
+    - pytest -v -s entrypoints/openai/test_tensorizer_entrypoint.py
 
 - label: Benchmarks # 11min
   timeout_in_minutes: 20

.github/mergify.yml

@@ -274,7 +274,7 @@ pull_request_rules:
       - files~=^vllm/model_executor/model_loader/tensorizer.py
       - files~=^vllm/model_executor/model_loader/tensorizer_loader.py
       - files~=^tests/entrypoints/openai/test_tensorizer_entrypoint.py
-      - files~=^tests/tensorizer_loader/
+      - files~=^tests/model_executor/model_loader/tensorizer_loader/
   actions:
     assign:
       users:

docker/Dockerfile

@@ -546,7 +546,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
     else \
         BITSANDBYTES_VERSION="0.46.1"; \
     fi; \
-    uv pip install --system accelerate hf_transfer modelscope "bitsandbytes>=${BITSANDBYTES_VERSION}" 'timm>=1.0.17' boto3 runai-model-streamer runai-model-streamer[s3]
+    uv pip install --system accelerate hf_transfer modelscope "bitsandbytes>=${BITSANDBYTES_VERSION}" 'timm>=1.0.17' 'runai-model-streamer[s3]>=0.14.0'
 
 ENV VLLM_USAGE_SOURCE production-docker-image
 

requirements/nightly_torch_test.txt

@@ -43,7 +43,6 @@ tritonclient==2.51.0
 numba == 0.60.0; python_version == '3.9' # v0.61 doesn't support Python 3.9. Required for N-gram speculative decoding
 numba == 0.61.2; python_version > '3.9'
 numpy
-runai-model-streamer==0.11.0
-runai-model-streamer-s3==0.11.0
+runai-model-streamer[s3]==0.14.0
 fastsafetensors>=0.1.10
 pydantic>=2.10 # 2.9 leads to error on python 3.10

requirements/rocm.txt

@@ -5,8 +5,6 @@ numba == 0.60.0; python_version == '3.9' # v0.61 doesn't support Python 3.9. Req
 numba == 0.61.2; python_version > '3.9'
 
 # Dependencies for AMD GPUs
-boto3
-botocore
 datasets
 ray[cgraph]>=2.48.0 # Ray Compiled Graph, required for pipeline parallelism in V1.
 peft
@@ -15,7 +13,6 @@ tensorizer==2.10.1
 packaging>=24.2
 setuptools>=77.0.3,<80.0.0
 setuptools-scm>=8
-runai-model-streamer==0.11.0
-runai-model-streamer-s3==0.11.0
+runai-model-streamer[s3]==0.14.0
 conch-triton-kernels==1.2.1
 timm>=1.0.17

requirements/test.in

@@ -51,8 +51,7 @@ tritonclient==2.51.0
 numba == 0.60.0; python_version == '3.9' # v0.61 doesn't support Python 3.9. Required for N-gram speculative decoding
 numba == 0.61.2; python_version > '3.9'
 numpy
-runai-model-streamer==0.11.0
-runai-model-streamer-s3==0.11.0
+runai-model-streamer[s3]==0.14.0
 fastsafetensors>=0.1.10
 pydantic>=2.10 # 2.9 leads to error on python 3.10
 decord==0.6.0

requirements/test.txt

@@ -72,7 +72,9 @@ blobfile==3.0.0
 bm25s==0.2.13
     # via mteb
 boto3==1.35.57
-    # via tensorizer
+    # via
+    #   runai-model-streamer-s3
+    #   tensorizer
 botocore==1.35.57
     # via
     #   boto3
@@ -925,10 +927,10 @@ rsa==4.9.1
     # via google-auth
 rtree==1.4.0
     # via torchgeo
-runai-model-streamer==0.11.0
-    # via -r requirements/test.in
-runai-model-streamer-s3==0.11.0
+runai-model-streamer==0.14.0
     # via -r requirements/test.in
+runai-model-streamer-s3==0.14.0
+    # via runai-model-streamer
 s3transfer==0.10.3
     # via boto3
 sacrebleu==2.4.3

setup.py

@@ -654,10 +654,7 @@ setup(
         "bench": ["pandas", "datasets"],
         "tensorizer": ["tensorizer==2.10.1"],
         "fastsafetensors": ["fastsafetensors >= 0.1.10"],
-        "runai": [
-            "runai-model-streamer >= 0.14.0", "runai-model-streamer-gcs",
-            "google-cloud-storage", "runai-model-streamer-s3", "boto3"
-        ],
+        "runai": ["runai-model-streamer[s3,gcs] >= 0.14.0"],
         "audio": ["librosa", "soundfile",
                   "mistral_common[audio]"],  # Required for audio processing
         "video": [],  # Kept for backwards compatibility

tests/model_executor/conftest.py

@@ -1,52 +0,0 @@
-# SPDX-License-Identifier: Apache-2.0
-# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-
-import pytest
-
-
-@pytest.fixture
-def sample_regex():
-    return (r"((25[0-5]|(2[0-4]|1\d|[1-9]|)\d)\.){3}"
-            r"(25[0-5]|(2[0-4]|1\d|[1-9]|)\d)")
-
-
-@pytest.fixture
-def sample_json_schema():
-    return {
-        "type": "object",
-        "properties": {
-            "name": {
-                "type": "string"
-            },
-            "age": {
-                "type": "integer"
-            },
-            "skills": {
-                "type": "array",
-                "items": {
-                    "type": "string",
-                    "maxLength": 10
-                },
-                "minItems": 3
-            },
-            "work_history": {
-                "type": "array",
-                "items": {
-                    "type": "object",
-                    "properties": {
-                        "company": {
-                            "type": "string"
-                        },
-                        "duration": {
-                            "type": "number"
-                        },
-                        "position": {
-                            "type": "string"
-                        }
-                    },
-                    "required": ["company", "position"]
-                }
-            }
-        },
-        "required": ["name", "age", "skills", "work_history"]
-    }

tests/model_executor/model_loader/tensorizer_loader/test_tensorizer.py

@@ -14,6 +14,7 @@ import pytest
 import torch
 
 import vllm.model_executor.model_loader.tensorizer
+from tests.utils import VLLM_PATH, RemoteOpenAIServer
 from vllm import LLM, SamplingParams
 from vllm.engine.arg_utils import EngineArgs
 # yapf: disable
@@ -27,7 +28,6 @@ from vllm.model_executor.model_loader.tensorizer_loader import (
 # yapf: enable
 from vllm.utils import PlaceholderModule
 
-from ..utils import VLLM_PATH, RemoteOpenAIServer
 from .conftest import DummyExecutor, assert_from_collective_rpc
 
 try:

vllm/model_executor/model_loader/weight_utils.py

@@ -639,6 +639,19 @@ def runai_safetensors_weights_iterator(
         yield from tensor_iter
 
 
+def _init_loader(
+    pg: torch.distributed.ProcessGroup,
+    device: torch.device,
+    f_list: list[str],
+    *,
+    nogds: bool = False,
+):
+    loader = SafeTensorsFileLoader(pg, device, nogds=nogds)
+    rank_file_map = {i: [f] for i, f in enumerate(f_list)}
+    loader.add_filenames(rank_file_map)
+    return loader
+
+
 def fastsafetensors_weights_iterator(
     hf_weights_files: list[str],
     use_tqdm_on_load: bool,
@@ -656,17 +669,31 @@ def fastsafetensors_weights_iterator(
         for i in range(0, len(hf_weights_files), pg.size())
     ]
 
+    nogds = False
+
     for f_list in tqdm(
             weight_files_sub_lists,
             desc="Loading safetensors using Fastsafetensor loader",
             disable=not enable_tqdm(use_tqdm_on_load),
             bar_format=_BAR_FORMAT,
     ):
-        loader = SafeTensorsFileLoader(pg, device)
-        rank_file_map = {i: [f] for i, f in enumerate(f_list)}
-        loader.add_filenames(rank_file_map)
+        loader = _init_loader(pg, device, f_list, nogds=nogds)
         try:
-            fb = loader.copy_files_to_device()
+            try:
+                fb = loader.copy_files_to_device()
+            except RuntimeError as e:
+                if "gds" not in str(e):
+                    raise
+
+                loader.close()
+                nogds = True
+                logger.warning_once(
+                    "GDS not enabled, setting `nogds=True`.\n"
+                    "For more information, see: https://github.com/foundation-model-stack/fastsafetensors?tab=readme-ov-file#basic-api-usages"
+                )
+                loader = _init_loader(pg, device, f_list, nogds=nogds)
+                fb = loader.copy_files_to_device()
+
             try:
                 keys = list(fb.key_to_rank_lidx.keys())
                 for k in keys: