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 6, 19:15 UTC
Update Flashinfer to 0.6.6 (#36768)
Signed-off-by: Dimitrios Bariamis <12195802+dbari@users.noreply.github.com> Co-authored-by: Dimitrios Bariamis <12195802+dbari@users.noreply.github.com>
Dimitrios Bariamis committed
Mar 12, 2026 at 18:19 UTC
cc16b24b17986c1983e3f30c0438e52b0328f9bd
5 files changed
+6
-6
docker/Dockerfile
+1
-1
@@ -586,7 +586,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
586
# This is ~1.1GB and only changes when FlashInfer version bumps
587
# https://docs.flashinfer.ai/installation.html
588
# From versions.json: .flashinfer.version
589
-ARG FLASHINFER_VERSION=0.6.4
589
+ARG FLASHINFER_VERSION=0.6.6
590
RUN --mount=type=cache,target=/root/.cache/uv \
591
uv pip install --system flashinfer-cubin==${FLASHINFER_VERSION} \
592
&& uv pip install --system flashinfer-jit-cache==${FLASHINFER_VERSION} \
docker/Dockerfile.nightly_torch
+2
-2
@@ -217,13 +217,13 @@ RUN pip install setuptools==75.6.0 packaging==23.2 ninja==1.11.1.3 build==1.2.2.
217
218
219
# build flashinfer for torch nightly from source around 10 mins
220
-# release version: v0.6.4
220
+# release version: v0.6.6
221
# todo(elainewy): cache flashinfer build result for faster build
222
ENV CCACHE_DIR=/root/.cache/ccache
223
RUN --mount=type=cache,target=/root/.cache/ccache \
224
--mount=type=cache,target=/root/.cache/uv \
225
echo "git clone flashinfer..." \
226
- && git clone --depth 1 --branch v0.6.4 --recursive https://github.com/flashinfer-ai/flashinfer.git \
226
+ && git clone --depth 1 --branch v0.6.6 --recursive https://github.com/flashinfer-ai/flashinfer.git \
227
&& cd flashinfer \
228
&& git submodule update --init --recursive \
229
&& echo "finish git clone flashinfer..." \
docker/versions.json
+1
-1
@@ -65,7 +65,7 @@
65
"default": "true"
66
},
67
"FLASHINFER_VERSION": {
68
- "default": "0.6.4"
68
+ "default": "0.6.6"
69
},
70
"GDRCOPY_CUDA_VERSION": {
71
"default": "12.8"
requirements/cuda.txt
+1
-1
@@ -9,7 +9,7 @@ torchaudio==2.10.0
9
# These must be updated alongside torch
10
torchvision==0.25.0 # Required for phi3v processor. See https://github.com/pytorch/vision?tab=readme-ov-file#installation for corresponding version
11
# FlashInfer should be updated together with the Dockerfile
12
-flashinfer-python==0.6.4
12
+flashinfer-python==0.6.6
13
# Cap nvidia-cudnn-frontend (transitive dep of flashinfer) due to
14
# breaking changes in 1.19.0
15
nvidia-cudnn-frontend>=1.13.0,<1.19.0
tools/pre_commit/update-dockerfile-graph.sh
+1
-1
@@ -41,7 +41,7 @@ if printf '%s\n' "${FILES[@]}" | grep -q "^docker/Dockerfile$"; then
41
--rm \
42
--user "$(id -u):$(id -g)" \
43
--workdir /workspace \
44
- --volume "$(pwd)":/workspace \
44
+ --volume "$(pwd -P)":/workspace \
45
ghcr.io/patrickhoefler/dockerfilegraph:alpine \
46
--output png \
47
--dpi 200 \