[ROCm][CI] Resolved nvidia package deps issue (#39421)

Signed-off-by: Andreas Karatzas <akaratza@amd.com>

.pre-commit-config.yaml

@@ -59,21 +59,54 @@ repos:
         --no-emit-package, cuda-pathfinder,
         --no-emit-package, cuda-toolkit,
         --no-emit-package, cupy-cuda12x,
+        # nvidia packages (unsuffixed / unified naming)
         --no-emit-package, nvidia-cublas,
         --no-emit-package, nvidia-cuda-cupti,
         --no-emit-package, nvidia-cuda-nvrtc,
         --no-emit-package, nvidia-cuda-runtime,
-        --no-emit-package, nvidia-cudnn-cu13,
+        --no-emit-package, nvidia-cudnn,
         --no-emit-package, nvidia-cufft,
         --no-emit-package, nvidia-cufile,
         --no-emit-package, nvidia-curand,
         --no-emit-package, nvidia-cusolver,
         --no-emit-package, nvidia-cusparse,
+        --no-emit-package, nvidia-cusparselt,
+        --no-emit-package, nvidia-nccl,
+        --no-emit-package, nvidia-nvjitlink,
+        --no-emit-package, nvidia-nvshmem,
+        --no-emit-package, nvidia-nvtx,
+        # nvidia cu12 packages
+        --no-emit-package, nvidia-cublas-cu12,
+        --no-emit-package, nvidia-cuda-cupti-cu12,
+        --no-emit-package, nvidia-cuda-nvrtc-cu12,
+        --no-emit-package, nvidia-cuda-runtime-cu12,
+        --no-emit-package, nvidia-cudnn-cu12,
+        --no-emit-package, nvidia-cufft-cu12,
+        --no-emit-package, nvidia-cufile-cu12,
+        --no-emit-package, nvidia-curand-cu12,
+        --no-emit-package, nvidia-cusolver-cu12,
+        --no-emit-package, nvidia-cusparse-cu12,
+        --no-emit-package, nvidia-cusparselt-cu12,
+        --no-emit-package, nvidia-nccl-cu12,
+        --no-emit-package, nvidia-nvjitlink-cu12,
+        --no-emit-package, nvidia-nvshmem-cu12,
+        --no-emit-package, nvidia-nvtx-cu12,
+        # nvidia cu13 packages
+        --no-emit-package, nvidia-cublas-cu13,
+        --no-emit-package, nvidia-cuda-cupti-cu13,
+        --no-emit-package, nvidia-cuda-nvrtc-cu13,
+        --no-emit-package, nvidia-cuda-runtime-cu13,
+        --no-emit-package, nvidia-cudnn-cu13,
+        --no-emit-package, nvidia-cufft-cu13,
+        --no-emit-package, nvidia-cufile-cu13,
+        --no-emit-package, nvidia-curand-cu13,
+        --no-emit-package, nvidia-cusolver-cu13,
+        --no-emit-package, nvidia-cusparse-cu13,
         --no-emit-package, nvidia-cusparselt-cu13,
         --no-emit-package, nvidia-nccl-cu13,
-        --no-emit-package, nvidia-nvjitlink,
+        --no-emit-package, nvidia-nvjitlink-cu13,
         --no-emit-package, nvidia-nvshmem-cu13,
-        --no-emit-package, nvidia-nvtx,
+        --no-emit-package, nvidia-nvtx-cu13,
       ]
       files: ^requirements/rocm-test\.(in|txt)$
 - repo: local

docker/Dockerfile.rocm

@@ -333,10 +333,10 @@ RUN --mount=type=bind,from=export_vllm,src=/,target=/install \
     && pip uninstall -y vllm \
     && uv pip install --system *.whl
 
-# Verify that PyTorch is the ROCm build, not CUDA
-RUN python3 -c "import torch; assert torch.version.hip is not None, \
-    f'Expected ROCm PyTorch but got CUDA (torch.version.cuda={torch.version.cuda}, torch.version.hip={torch.version.hip})'; \
-    print(f'Verified: PyTorch {torch.__version__} with ROCm (HIP {torch.version.hip})')"
+# Persist the built wheel in the image so python_only_compile_rocm.sh can
+# reinstall it after removing compilers. The bind-mounted /install contents
+# above are not available once that RUN step completes.
+COPY --from=export_vllm /*.whl /opt/vllm-wheels/
 
 # Install RIXL wheel
 RUN --mount=type=bind,from=build_rixl,src=/app/install,target=/rixl_install \

requirements/rocm-test.txt

 # This file was autogenerated by uv via the following command:
-#    uv pip compile requirements/rocm-test.in -o requirements/rocm-test.txt --index-strategy unsafe-best-match -c requirements/rocm.txt --python-platform x86_64-manylinux_2_28 --python-version 3.12 --no-emit-package torch --no-emit-package torchvision --no-emit-package torchaudio --no-emit-package triton --no-emit-package cuda-bindings --no-emit-package cuda-pathfinder --no-emit-package cuda-toolkit --no-emit-package cupy-cuda12x --no-emit-package nvidia-cublas --no-emit-package nvidia-cuda-cupti --no-emit-package nvidia-cuda-nvrtc --no-emit-package nvidia-cuda-runtime --no-emit-package nvidia-cudnn-cu13 --no-emit-package nvidia-cufft --no-emit-package nvidia-cufile --no-emit-package nvidia-curand --no-emit-package nvidia-cusolver --no-emit-package nvidia-cusparse --no-emit-package nvidia-cusparselt-cu13 --no-emit-package nvidia-nccl-cu13 --no-emit-package nvidia-nvjitlink --no-emit-package nvidia-nvshmem-cu13 --no-emit-package nvidia-nvtx
+#    uv pip compile requirements/rocm-test.in -o requirements/rocm-test.txt --index-strategy unsafe-best-match -c requirements/rocm.txt --python-platform x86_64-manylinux_2_28 --python-version 3.12 --no-emit-package torch --no-emit-package torchvision --no-emit-package torchaudio --no-emit-package triton --no-emit-package cuda-bindings --no-emit-package cuda-pathfinder --no-emit-package cuda-toolkit --no-emit-package cupy-cuda12x --no-emit-package nvidia-cublas --no-emit-package nvidia-cuda-cupti --no-emit-package nvidia-cuda-nvrtc --no-emit-package nvidia-cuda-runtime --no-emit-package nvidia-cudnn --no-emit-package nvidia-cufft --no-emit-package nvidia-cufile --no-emit-package nvidia-curand --no-emit-package nvidia-cusolver --no-emit-package nvidia-cusparse --no-emit-package nvidia-cusparselt --no-emit-package nvidia-nccl --no-emit-package nvidia-nvjitlink --no-emit-package nvidia-nvshmem --no-emit-package nvidia-nvtx --no-emit-package nvidia-cublas-cu12 --no-emit-package nvidia-cuda-cupti-cu12 --no-emit-package nvidia-cuda-nvrtc-cu12 --no-emit-package nvidia-cuda-runtime-cu12 --no-emit-package nvidia-cudnn-cu12 --no-emit-package nvidia-cufft-cu12 --no-emit-package nvidia-cufile-cu12 --no-emit-package nvidia-curand-cu12 --no-emit-package nvidia-cusolver-cu12 --no-emit-package nvidia-cusparse-cu12 --no-emit-package nvidia-cusparselt-cu12 --no-emit-package nvidia-nccl-cu12 --no-emit-package nvidia-nvjitlink-cu12 --no-emit-package nvidia-nvshmem-cu12 --no-emit-package nvidia-nvtx-cu12 --no-emit-package nvidia-cublas-cu13 --no-emit-package nvidia-cuda-cupti-cu13 --no-emit-package nvidia-cuda-nvrtc-cu13 --no-emit-package nvidia-cuda-runtime-cu13 --no-emit-package nvidia-cudnn-cu13 --no-emit-package nvidia-cufft-cu13 --no-emit-package nvidia-cufile-cu13 --no-emit-package nvidia-curand-cu13 --no-emit-package nvidia-cusolver-cu13 --no-emit-package nvidia-cusparse-cu13 --no-emit-package nvidia-cusparselt-cu13 --no-emit-package nvidia-nccl-cu13 --no-emit-package nvidia-nvjitlink-cu13 --no-emit-package nvidia-nvshmem-cu13 --no-emit-package nvidia-nvtx-cu13
 absl-py==2.4.0
     # via
     #   rouge-score
@@ -1634,14 +1634,14 @@ zstandard==0.25.0
 # nvidia-cuda-cupti
 # nvidia-cuda-nvrtc
 # nvidia-cuda-runtime
-# nvidia-cudnn-cu13
 # nvidia-cufft
 # nvidia-cufile
 # nvidia-curand
 # nvidia-cusolver
 # nvidia-cusparse
+# nvidia-nvjitlink
+# nvidia-nvtx
+# nvidia-cudnn-cu13
 # nvidia-cusparselt-cu13
 # nvidia-nccl-cu13
-# nvidia-nvjitlink
 # nvidia-nvshmem-cu13
-# nvidia-nvtx