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

changed arch to 12.1a again

Browse Source
This commit is contained in:
Eugene Rakhmatulin
2026-02-08 14:18:12 -08:00
parent 5bf422a2ca
commit d845cd0401
2 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
Dockerfile.mxfp4
View File

@@ -67,7 +67,7 @@ ENV CMAKE_CUDA_COMPILER_LAUNCHER=ccache
WORKDIR $VLLM_BASE_DIR
# 2. Set Environment Variables
ENV TORCH_CUDA_ARCH_LIST="12.1f"
ENV TORCH_CUDA_ARCH_LIST="12.1a"
ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
# --- CACHE BUSTER ---
@@ -88,7 +88,7 @@ RUN --mount=type=cache,id=uv-cache,target=/root/.cache/uv \
# =========================================================
FROM base AS builder
ENV FLASHINFER_CUDA_ARCH_LIST="12.1f"
ENV FLASHINFER_CUDA_ARCH_LIST="12.1a"
WORKDIR $VLLM_BASE_DIR
@@ -261,8 +261,8 @@ RUN --mount=type=bind,from=builder,source=/workspace/wheels,target=/mount/wheels
uv pip install /mount/wheels/*.whl
# Setup Env for Runtime
ENV TORCH_CUDA_ARCH_LIST="12.1f"
ENV FLASHINFER_CUDA_ARCH_LIST="12.1f"
ENV TORCH_CUDA_ARCH_LIST="12.1a"
ENV FLASHINFER_CUDA_ARCH_LIST="12.1a"
ENV TRITON_PTXAS_PATH=/usr/local/cuda/bin/ptxas
ENV TIKTOKEN_ENCODINGS_BASE=$VLLM_BASE_DIR/tiktoken_encodings
ENV PATH=$VLLM_BASE_DIR:$PATH