tmp: pin modelscope

Signed-off-by: sirutBuasai <sirutbuasai27@outlook.com>

Author: sirutBuasai

docker/vllm/Dockerfile.amzn2023

@@ -199,7 +199,9 @@ RUN --mount=type=cache,target=/root/.cache/uv uv pip install \
   --extra-index-url https://flashinfer.ai/whl/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
 
 # Install serving extras
-RUN --mount=type=cache,target=/root/.cache/uv uv pip install accelerate modelscope \
+# modelscope<1.38: 1.38+ removed the `revision` kwarg from LegacyHubApi.get_model_files(),
+# breaking vLLM's repo_utils.py. Unpin after upstream vLLM adapts to the new API.
+RUN --mount=type=cache,target=/root/.cache/uv uv pip install accelerate "modelscope<1.38" \
   "bitsandbytes>=0.46.1" "timm>=1.0.17" "runai-model-streamer[s3,gcs,azure]>=0.15.7"
 
 # Audio deps for voxtral / ASR serving (mistral_common is an optional import in vLLM 0.24.0)

test/vllm/scripts/vllm_regression_test.sh

@@ -9,5 +9,7 @@ fi
 
 # Regression Test # 7min
 cd vllm_source/tests
-uv pip install $UV_FLAGS modelscope
+# modelscope<1.38: 1.38+ removed `revision` kwarg from LegacyHubApi.get_model_files(),
+# breaking vLLM's repo_utils.py. Unpin after upstream vLLM adapts to the new API.
+uv pip install $UV_FLAGS "modelscope<1.38"
 pytest -v -s test_regression.py