kvcache-ai/ktransformers
1
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-03-14 18:37:23 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix cuda wheel build (#1766)

Browse Source
This commit is contained in:
Jianwei Dong
2025-12-29 12:42:06 +08:00
committed by GitHub
parent 7c127d9fd0
commit 4b235cdaa4
3 changed files with 33 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
kt-kernel/README.md
View File

@@ -48,13 +48,7 @@ High-performance kernel operations for KTransformers, featuring CPU-optimized Mo
Install the latest CPU-only version:
```bash
pip install "kt-kernel==0.5.0+cpu"
```
Or let pip auto-select the latest CPU version:
```bash
pip install kt-kernel # Defaults to CPU version
pip install kt-kernel
```
> **Note**: Check the [latest version on PyPI](https://pypi.org/project/kt-kernel/#history)
@@ -75,7 +69,7 @@ pip install kt-kernel # Defaults to CPU version
For NVIDIA GPU-accelerated inference:
```bash
pip install "kt-kernel==0.5.0+cuda118"
pip install kt-kernel-cuda
```
**Features:**

28
kt-kernel/setup.py
View File

@@ -698,31 +698,33 @@ if _version_file.exists():
else:
_base_version = "0.5.0"
# Auto-detect version suffix based on build type
# Determine package name and version based on build type
# PyPI doesn't allow local version identifiers (+suffix), so we use separate package names
if "CPUINFER_VERSION" in os.environ:
# User explicitly set version (e.g., for testing)
VERSION = os.environ["CPUINFER_VERSION"]
print(f"-- Explicit version: {VERSION}")
else:
# Auto-detect suffix based on CUDA usage
cuda_enabled = _env_get_bool("CPUINFER_USE_CUDA", False)
VERSION = _base_version
if cuda_enabled:
# CUDA build: add +cuda118 suffix
# (CUDA 11.8 is the build toolkit version for compatibility with 11.8+ and 12.x)
VERSION = f"{_base_version}+cuda118"
print(f"-- CUDA wheel version: {VERSION}")
else:
# CPU-only build: add +cpu suffix
VERSION = f"{_base_version}+cpu"
print(f"-- CPU wheel version: {VERSION}")
# Determine package name based on CUDA usage
cuda_enabled = _env_get_bool("CPUINFER_USE_CUDA", False)
if cuda_enabled:
# CUDA build: use kt-kernel-cuda package name
# Compatible with CUDA 11.8+ and 12.x drivers
PACKAGE_NAME = "kt-kernel-cuda"
print(f"-- CUDA wheel: {PACKAGE_NAME} version {VERSION}")
else:
# CPU-only build: use kt-kernel package name
PACKAGE_NAME = "kt-kernel"
print(f"-- CPU wheel: {PACKAGE_NAME} version {VERSION}")
################################################################################
# Setup
################################################################################
setup(
name="kt-kernel",
name=PACKAGE_NAME,
version=VERSION,
description="KT-Kernel: High-performance kernel operations for KTransformers (AMX/AVX/KML optimizations)",
author="kvcache-ai",