From c1076839c9f14a51c0eb963ca8ae12c2de3c0f63 Mon Sep 17 00:00:00 2001 From: Ting SUN Date: Fri, 12 Jun 2026 11:21:46 +0800 Subject: [PATCH] [Bugfix][Model] Pass revision by name in Run:ai and bitsandbytes index downloads (#45308) Signed-off-by: Ting Sun --- .../test_runai_model_streamer_loader.py | 25 +++++++++++++++++ .../models/quantization/test_bitsandbytes.py | 28 +++++++++++++++++++ .../model_loader/bitsandbytes_loader.py | 4 +-- .../model_loader/runai_streamer_loader.py | 5 +++- 4 files changed, 59 insertions(+), 3 deletions(-) diff --git a/tests/model_executor/model_loader/runai_streamer_loader/test_runai_model_streamer_loader.py b/tests/model_executor/model_loader/runai_streamer_loader/test_runai_model_streamer_loader.py index c7158dae53..82c0f8813e 100644 --- a/tests/model_executor/model_loader/runai_streamer_loader/test_runai_model_streamer_loader.py +++ b/tests/model_executor/model_loader/runai_streamer_loader/test_runai_model_streamer_loader.py @@ -1,11 +1,15 @@ # SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project +import types +from unittest.mock import patch + import pytest from vllm import SamplingParams from vllm.config.load import LoadConfig from vllm.model_executor.model_loader import get_model_loader +from vllm.model_executor.model_loader import runai_streamer_loader as rsl load_format = "runai_streamer" test_model = "openai-community/gpt2" @@ -53,3 +57,24 @@ def test_runai_model_loader_download_files_gcs( with vllm_runner(test_gcs_model, load_format=load_format) as llm: deserialized_outputs = llm.generate(prompts, sampling_params) assert deserialized_outputs + + +def test_runai_passes_revision_by_name(): + # revision must reach download_safetensors_index_file_from_hf as the + # ``revision`` keyword, not the positional ``subfolder`` slot. + fake_self = types.SimpleNamespace( + load_config=types.SimpleNamespace(download_dir="/cache", ignore_patterns=[]) + ) + with ( + patch.object(rsl, "is_runai_obj_uri", return_value=False), + patch.object(rsl, "download_weights_from_hf", return_value="/folder"), + patch.object( + rsl, "list_safetensors", return_value=["/folder/model.safetensors"] + ), + patch.object(rsl, "download_safetensors_index_file_from_hf") as mock_idx, + ): + rsl.RunaiModelStreamerLoader._prepare_weights(fake_self, "org/model", "myrev") + + mock_idx.assert_called_once() + assert mock_idx.call_args.kwargs.get("revision") == "myrev" + assert "myrev" not in mock_idx.call_args.args diff --git a/tests/models/quantization/test_bitsandbytes.py b/tests/models/quantization/test_bitsandbytes.py index d6f2b86c7a..03c19b0bf6 100644 --- a/tests/models/quantization/test_bitsandbytes.py +++ b/tests/models/quantization/test_bitsandbytes.py @@ -5,12 +5,16 @@ Run `pytest tests/quantization/test_bitsandbytes.py`. """ +import types +from unittest.mock import MagicMock, patch + import pytest from packaging.version import Version from transformers import BitsAndBytesConfig from transformers import __version__ as TRANSFORMERS_VERSION from tests.quantization.utils import is_quant_method_supported +from vllm.model_executor.model_loader import bitsandbytes_loader as bnb from vllm.platforms import current_platform from ...utils import compare_two_settings, multi_gpu_test @@ -300,3 +304,27 @@ def validate_generated_texts( f"HF Output: '{hf_str}'\n" f"vLLM Output: '{vllm_str}'" ) + + +def test_bitsandbytes_passes_revision_by_name(): + # revision must reach download_safetensors_index_file_from_hf as the + # ``revision`` keyword, not a positional slot. + fake_self = types.SimpleNamespace( + load_config=types.SimpleNamespace(download_dir="/cache"), + _get_weight_files=MagicMock( + return_value=("/folder", ["/folder/model.safetensors"], "*.safetensors") + ), + ) + with ( + patch.object(bnb, "download_safetensors_index_file_from_hf") as mock_idx, + patch.object( + bnb, + "filter_duplicate_safetensors_files", + return_value=["/folder/model.safetensors"], + ), + ): + bnb.BitsAndBytesModelLoader._prepare_weights(fake_self, "org/model", "myrev") + + mock_idx.assert_called_once() + assert mock_idx.call_args.kwargs.get("revision") == "myrev" + assert "myrev" not in mock_idx.call_args.args diff --git a/vllm/model_executor/model_loader/bitsandbytes_loader.py b/vllm/model_executor/model_loader/bitsandbytes_loader.py index d10f3bfcbe..064a74023a 100644 --- a/vllm/model_executor/model_loader/bitsandbytes_loader.py +++ b/vllm/model_executor/model_loader/bitsandbytes_loader.py @@ -140,8 +140,8 @@ class BitsAndBytesModelLoader(BaseModelLoader): download_safetensors_index_file_from_hf( model_name_or_path, index_file, - self.load_config.download_dir, - revision, + cache_dir=self.load_config.download_dir, + revision=revision, ) hf_weights_files = filter_duplicate_safetensors_files( hf_weights_files, hf_folder, index_file diff --git a/vllm/model_executor/model_loader/runai_streamer_loader.py b/vllm/model_executor/model_loader/runai_streamer_loader.py index 47c3c99b19..0df1422791 100644 --- a/vllm/model_executor/model_loader/runai_streamer_loader.py +++ b/vllm/model_executor/model_loader/runai_streamer_loader.py @@ -70,7 +70,10 @@ class RunaiModelStreamerLoader(BaseModelLoader): if not is_local and not is_object_storage_path: download_safetensors_index_file_from_hf( - model_name_or_path, index_file, self.load_config.download_dir, revision + model_name_or_path, + index_file, + cache_dir=self.load_config.download_dir, + revision=revision, ) if not hf_weights_files: