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 14, 05:30 UTC
[CI] Install pre-release version of `apache-tvm-ffi` for `flashinfer` (#27262)
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Harry Mellor committed
Oct 21, 2025 at 15:23 UTC
bd66b8529bb00821bf14565e8b7663cbdc286d91
1 file changed
+6
-2
docker/Dockerfile
+6
-2
@@ -132,7 +132,9 @@ WORKDIR /workspace
132
COPY requirements/common.txt requirements/common.txt
133
COPY requirements/cuda.txt requirements/cuda.txt
134
RUN --mount=type=cache,target=/root/.cache/uv \
135
- uv pip install --python /opt/venv/bin/python3 -r requirements/cuda.txt \
135
+ # TODO: remove apache-tvm-ffi once FlashInfer is fixed https://github.com/flashinfer-ai/flashinfer/issues/1962
136
+ uv pip install --python /opt/venv/bin/python3 --pre apache-tvm-ffi==0.1.0b15 \
137
+ && uv pip install --python /opt/venv/bin/python3 -r requirements/cuda.txt \
138
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
139
140
# cuda arch list used by torch
@@ -353,7 +355,9 @@ RUN --mount=type=cache,target=/root/.cache/uv \
355
# Install vllm wheel first, so that torch etc will be installed.
356
RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
357
--mount=type=cache,target=/root/.cache/uv \
356
- uv pip install --system dist/*.whl --verbose \
358
+ # TODO: remove apache-tvm-ffi once FlashInfer is fixed https://github.com/flashinfer-ai/flashinfer/issues/1962
359
+ uv pip install --system --pre apache-tvm-ffi==0.1.0b15 \
360
+ && uv pip install --system dist/*.whl --verbose \
361
--extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.')
362
363
# Install FlashInfer pre-compiled kernel cache and binaries