software-engineering/spark-vllm-docker
2
0
Fork 0
Code Issues Actions Packages Projects Releases Wiki Activity

Fixes

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-03-20 23:51:21 -07:00
parent 8caebe3155
commit 8385506c5e
1 changed files with 8 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
Dockerfile
View File

@@ -43,6 +43,11 @@ RUN apt update && \
&& rm -rf /var/lib/apt/lists/* \ && rm -rf /var/lib/apt/lists/* \
&& pip install uv && pip install uv
# Additional deps
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install torch torchvision torchaudio triton --index-url https://download.pytorch.org/whl/nightly/cu130 && \
uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2" filelock pynvml requests tqdm
# Configure Ccache for CUDA/C++ # Configure Ccache for CUDA/C++
ENV PATH=/usr/lib/ccache:$PATH ENV PATH=/usr/lib/ccache:$PATH
ENV CCACHE_DIR=/root/.ccache ENV CCACHE_DIR=/root/.ccache
@@ -76,9 +81,6 @@ ARG FLASHINFER_REF=main
# Change this argument to force a re-download of FlashInfer # Change this argument to force a re-download of FlashInfer
ARG CACHEBUST_FLASHINFER=1 ARG CACHEBUST_FLASHINFER=1
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2"
# Smart Git Clone (Fetch changes instead of full re-clone) # Smart Git Clone (Fetch changes instead of full re-clone)
RUN --mount=type=cache,id=repo-cache,target=/repo-cache \ RUN --mount=type=cache,id=repo-cache,target=/repo-cache \
cd /repo-cache && \ cd /repo-cache && \
@@ -135,10 +137,6 @@ ARG TORCH_CUDA_ARCH_LIST="12.1a"
ENV TORCH_CUDA_ARCH_LIST=${TORCH_CUDA_ARCH_LIST} ENV TORCH_CUDA_ARCH_LIST=${TORCH_CUDA_ARCH_LIST}
WORKDIR $VLLM_BASE_DIR WORKDIR $VLLM_BASE_DIR
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130 && \
uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2" triton
# --- VLLM SOURCE CACHE BUSTER --- # --- VLLM SOURCE CACHE BUSTER ---
ARG CACHEBUST_VLLM=1 ARG CACHEBUST_VLLM=1
@@ -256,10 +254,10 @@ RUN mkdir -p tiktoken_encodings && \
ARG PRE_TRANSFORMERS=0 ARG PRE_TRANSFORMERS=0
# Install dependencies # Install deps
RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
uv pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu130 && \ uv pip install torch torchvision torchaudio triton --index-url https://download.pytorch.org/whl/nightly/cu130 && \
uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2" triton uv pip install nvidia-nvshmem-cu13 "apache-tvm-ffi<0.2"
# Install wheels from host ./wheels/ (bind-mounted from build context — no layer bloat) # Install wheels from host ./wheels/ (bind-mounted from build context — no layer bloat)
# With --tf5: override vLLM's transformers<5 constraint to get transformers>=5 # With --tf5: override vLLM's transformers<5 constraint to get transformers>=5