x

Dockerfile

@@ -57,7 +57,7 @@ ENV PIP_NO_CACHE_DIR=1
 RUN printf 'source /opt/venv/bin/activate\n' > /etc/profile.d/venv.sh
 RUN python -m pip install --upgrade pip wheel packaging "setuptools<80.0.0"
 
-# 5. Install PyTorch (TheRock Nightly) and PyYAML
+# 5. Install PyTorch (TheRock Nightly)
 RUN proxychains python -m pip install \
   --index-url https://rocm.nightlies.amd.com/v2-staging/gfx120X-all/ \
   --pre torch torchaudio torchvision && \
@@ -66,6 +66,10 @@ RUN proxychains python -m pip install \
 # Flash-Attention
 WORKDIR /opt
 ENV FLASH_ATTENTION_TRITON_AMD_ENABLE="TRUE"
+ENV ROCM_HOME="/opt/rocm"
+ENV HIP_PATH="/opt/rocm"
+ENV PATH="/opt/rocm/bin:/opt/rocm/llvm/bin:$PATH"
+ENV LD_LIBRARY_PATH="/opt/rocm/lib:/opt/rocm/lib64:/opt/rocm/llvm/lib:$LD_LIBRARY_PATH"
 
 RUN proxychains git clone https://github.com/ROCm/flash-attention.git && \
   cd flash-attention && \
@@ -99,7 +103,7 @@ RUN echo "import sys, re" > patch_vllm.py && \
   echo "print('Successfully patched vLLM for R9700')" >> patch_vllm.py && \
   python patch_vllm.py
 
-# 7. Build vLLM
+# 7. Build vLLM (Wheel Method) with CLANG Host Compiler
 RUN python -m pip install --upgrade cmake ninja packaging wheel numpy "setuptools-scm>=8" "setuptools<80.0.0" scikit-build-core pybind11
 ENV ROCM_HOME="/opt/rocm"
 ENV HIP_PATH="/opt/rocm"