A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 09:45 UTC
Update Dockerfile to build for Blackwell (#18095)
Michael Goin committed
May 17, 2025 at 03:23 UTC
dcfe95234c1154abec9144266c66d137b12d4689
1 file changed
+3
-3
docker/Dockerfile
+3
-3
@@ -77,7 +77,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
77
# can be useful for both `dev` and `test`
78
# explicitly set the list to avoid issues with torch 2.2
79
# see https://github.com/pytorch/pytorch/pull/123243
80
-ARG torch_cuda_arch_list='7.0 7.5 8.0 8.6 8.9 9.0+PTX'
80
+ARG torch_cuda_arch_list='7.0 7.5 8.0 8.9 9.0 10.0+PTX'
81
ENV TORCH_CUDA_ARCH_LIST=${torch_cuda_arch_list}
82
# Override the arch list for flash-attn to reduce the binary size
83
ARG vllm_fa_cmake_gpu_arches='80-real;90-real'
@@ -257,8 +257,8 @@ RUN --mount=type=cache,target=/root/.cache/uv \
257
if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \
258
# uv pip install --system https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.4/flashinfer_python-0.2.4+cu124torch2.6-cp38-abi3-linux_x86_64.whl ; \
259
# TESTING: install FlashInfer from source to test 2.7.0 final RC
260
- FLASHINFER_ENABLE_AOT=1 TORCH_CUDA_ARCH_LIST='7.5 8.0 8.6 8.9 9.0+PTX' \
261
- uv pip install --system --no-build-isolation "git+https://github.com/flashinfer-ai/flashinfer@v0.2.4" ; \
260
+ FLASHINFER_ENABLE_AOT=1 TORCH_CUDA_ARCH_LIST='7.5 8.0 8.9 9.0 10.0+PTX' \
261
+ uv pip install --system --no-build-isolation "git+https://github.com/flashinfer-ai/flashinfer@e00e8cedbfcb220f328fd36aa8f529f869b01e6b" ; \
262
fi
263
COPY examples examples
264
COPY benchmarks benchmarks