vrr/kvcache-ai-ktransformers
2
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2025-09-10 15:29:39 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

bug fix

Browse source
This commit is contained in:
Alisehen 2025-05-15 10:01:51 +00:00
parent edd9efa49e
commit f3b1e36b6a
3 changed files with 11 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

8
install.sh
View file

@ -31,13 +31,6 @@ pip install -r ktransformers/server/requirements.txt
echo "Installing ktransformers" echo "Installing ktransformers"
KTRANSFORMERS_FORCE_BUILD=TRUE pip install -v . --no-build-isolation KTRANSFORMERS_FORCE_BUILD=TRUE pip install -v . --no-build-isolation
# XPU-specific fix for triton
if [[ "$DEV_BACKEND" == "xpu" ]]; then
echo "Replacing triton for XPU backend"
pip uninstall -y triton pytorch-triton-xpu || true
pip install pytorch-triton-xpu==3.3.0 --extra-index-url https://download.pytorch.org/whl/xpu
fi
if [[ "$DEV_BACKEND" == "cuda" ]]; then if [[ "$DEV_BACKEND" == "cuda" ]]; then
echo "Installing custom_flashinfer for CUDA backend" echo "Installing custom_flashinfer for CUDA backend"
pip install third_party/custom_flashinfer/ pip install third_party/custom_flashinfer/
@ -47,5 +40,4 @@ fi
# cp -a csrc/balance_serve/build/third_party/prometheus-cpp/lib/libprometheus-cpp-*.so* $SITE_PACKAGES/ # cp -a csrc/balance_serve/build/third_party/prometheus-cpp/lib/libprometheus-cpp-*.so* $SITE_PACKAGES/
# patchelf --set-rpath '$ORIGIN' $SITE_PACKAGES/sched_ext.cpython* # patchelf --set-rpath '$ORIGIN' $SITE_PACKAGES/sched_ext.cpython*
echo "Installation completed successfully" echo "Installation completed successfully"

1
requirements-local_chat.txt
View file

@ -7,4 +7,3 @@ cpufeature; sys_platform == 'win32' or sys_platform == 'Windows'
protobuf protobuf
tiktoken tiktoken
blobfile blobfile
triton>=3.2

10
setup.py
View file

@ -41,6 +41,15 @@ except ImportError:
MUSA_HOME=None MUSA_HOME=None
KTRANSFORMERS_BUILD_XPU = torch.xpu.is_available() KTRANSFORMERS_BUILD_XPU = torch.xpu.is_available()
# 检测 DEV_BACKEND 环境变量
dev_backend = os.environ.get("DEV_BACKEND", "").lower()
if dev_backend == "xpu":
triton_dep = [
"pytorch-triton-xpu==3.3.0"
]
else:
triton_dep = ["triton>=3.2"]
with_balance = os.environ.get("USE_BALANCE_SERVE", "0") == "1" with_balance = os.environ.get("USE_BALANCE_SERVE", "0") == "1"
class CpuInstructInfo: class CpuInstructInfo:
@ -659,6 +668,7 @@ else:
setup( setup(
name=VersionInfo.PACKAGE_NAME, name=VersionInfo.PACKAGE_NAME,
version=VersionInfo().get_package_version(), version=VersionInfo().get_package_version(),
install_requires=triton_dep,
cmdclass={"bdist_wheel":BuildWheelsCommand ,"build_ext": CMakeBuild}, cmdclass={"bdist_wheel":BuildWheelsCommand ,"build_ext": CMakeBuild},
ext_modules=ext_modules ext_modules=ext_modules
) )