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, 23:30 UTC
[ci/build] update flashinfer (#13323)
youkaichao committed
Feb 15, 2025 at 21:33 UTC
54ed913f3437f9107791e8ea410170b86c968d52
1 file changed
+5
-2
Dockerfile
+5
-2
@@ -195,19 +195,22 @@ RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist
195
--mount=type=cache,target=/root/.cache/pip \
196
python3 -m pip install dist/*.whl --verbose
197
198
-# How to build this FlashInfer wheel:
198
+# If we need to build FlashInfer wheel before its release:
199
# $ export FLASHINFER_ENABLE_AOT=1
200
# $ # Note we remove 7.0 from the arch list compared to the list below, since FlashInfer only supports sm75+
201
# $ export TORCH_CUDA_ARCH_LIST='7.5 8.0 8.6 8.9 9.0+PTX'
202
# $ git clone https://github.com/flashinfer-ai/flashinfer.git --recursive
203
# $ cd flashinfer
204
# $ git checkout 524304395bd1d8cd7d07db083859523fcaa246a4
205
+# $ rm -rf build
206
# $ python3 setup.py bdist_wheel --dist-dir=dist --verbose
207
+# $ ls dist
208
+# $ # upload the wheel to a public location, e.g. https://wheels.vllm.ai/flashinfer/524304395bd1d8cd7d07db083859523fcaa246a4/flashinfer_python-0.2.1.post1+cu124torch2.5-cp38-abi3-linux_x86_64.whl
209
210
RUN --mount=type=cache,target=/root/.cache/pip \
211
. /etc/environment && \
212
if [ "$TARGETPLATFORM" != "linux/arm64" ]; then \
210
- python3 -m pip install https://wheels.vllm.ai/flashinfer/524304395bd1d8cd7d07db083859523fcaa246a4/flashinfer_python-0.2.0.post1-cp${PYTHON_VERSION_STR}-cp${PYTHON_VERSION_STR}-linux_x86_64.whl; \
213
+ python3 -m pip install https://github.com/flashinfer-ai/flashinfer/releases/download/v0.2.1.post1/flashinfer_python-0.2.1.post1+cu124torch2.5-cp38-abi3-linux_x86_64.whl ; \
214
fi
215
COPY examples examples
216