minke.yu/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[Dep] Upgrade flashinfer to 0.6.14 (#29910)

Browse Source 
Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>
Co-authored-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
This commit is contained in:
Baizhou Zhang
2026-07-09 17:52:33 -07:00
committed by GitHub
co-authored by Brayden Zhong Mohammad Miadh Angkad
parent b2f9a95867
commit 2c6cd1ef41
6 changed files with 48 additions and 25 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/ci/cuda/ci_install_dependency.sh
+14 -1
View File
@@ -222,7 +222,8 @@ uninstall_stale_flashinfer() {
# - flashinfer-cubin: 150+ MB
# - flashinfer-jit-cache: 1.2+ GB
FLASHINFER_PYTHON_REQUIRED=$(grep -Po -m1 'flashinfer_python(\[[^]]+\])?==\K[0-9A-Za-z\.\-]+' python/pyproject.toml || echo "")
FLASHINFER_CUBIN_REQUIRED=$(grep -Po -m1 'flashinfer_cubin(\[[^]]+\])?==\K[0-9A-Za-z\.\-]+' python/pyproject.toml || echo "")
# flashinfer-cubin is no longer a pyproject dependency (installed explicitly below), tracks the same version as flashinfer_python
FLASHINFER_CUBIN_REQUIRED="$FLASHINFER_PYTHON_REQUIRED"
FLASHINFER_CUBIN_INSTALLED=$(pip show flashinfer-cubin 2>/dev/null | grep "^Version:" | awk '{print $2}' || echo "")
FLASHINFER_JIT_INSTALLED=$(pip show flashinfer-jit-cache 2>/dev/null | grep "^Version:" | awk '{print $2}' | sed 's/+.*//' || echo "")
FLASHINFER_JIT_CU_VERSION=$(pip show flashinfer-jit-cache 2>/dev/null | grep "^Version:" | awk '{print $2}' | sed -n 's/.*+//p' || echo "")
@@ -371,6 +372,17 @@ install_sglang_router() {
mark_step_done "${FUNCNAME[0]}"
}
install_flashinfer_cubin() {
if [ "$UNINSTALL_CUBIN" = false ]; then
echo "flashinfer-cubin==${FLASHINFER_CUBIN_REQUIRED} already installed, skipping install"
else
# flashinfer-cubin is CUDA-version-agnostic, unlike jit-cache, so its index-url has no cu${CU_VERSION} suffix
$PIP_CMD install "flashinfer-cubin==${FLASHINFER_CUBIN_REQUIRED}" --index-url https://flashinfer.ai/whl $PIP_INSTALL_SUFFIX
fi
mark_step_done "${FUNCNAME[0]}"
}
download_flashinfer_cache() {
UNINSTALL_JIT_CACHE="$UNINSTALL_JIT_CACHE" \
FLASHINFER_PYTHON_REQUIRED="$FLASHINFER_PYTHON_REQUIRED" \
@@ -583,6 +595,7 @@ main() {
fi
install_sglang_kernel
install_sglang_router
install_flashinfer_cubin
download_flashinfer_cache
force_reinstall_cutlass_dsl_libs_cu13
stabilize_flashinfer_jit_paths