kvcache-ai/ktransformers
1
0
Fork 0
mirror of https://github.com/kvcache-ai/ktransformers.git synced 2026-04-29 02:31:20 +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

31
.github/workflows/release-pypi.yml vendored
View File

@@ -282,36 +282,37 @@ jobs:
run: | run: |
echo "## 🎉 kt-kernel v${{ steps.get_version.outputs.VERSION }} Published to PyPI" >> $GITHUB_STEP_SUMMARY echo "## 🎉 kt-kernel v${{ steps.get_version.outputs.VERSION }} Published to PyPI" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "### Published Packages" >> $GITHUB_STEP_SUMMARY
echo "- **kt-kernel** (CPU-only)" >> $GITHUB_STEP_SUMMARY
echo "- **kt-kernel-cuda** (CUDA support)" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY
echo "Total wheels: $(ls -1 dist/*.whl | wc -l) (3 Python versions: 3.10, 3.11, 3.12)" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY
echo "### Installation" >> $GITHUB_STEP_SUMMARY echo "### Installation" >> $GITHUB_STEP_SUMMARY
echo '```bash' >> $GITHUB_STEP_SUMMARY echo '```bash' >> $GITHUB_STEP_SUMMARY
echo "# CPU version (AMX/AVX512/AVX2 multi-variant)" >> $GITHUB_STEP_SUMMARY
echo "pip install kt-kernel==${{ steps.get_version.outputs.VERSION }}" >> $GITHUB_STEP_SUMMARY echo "pip install kt-kernel==${{ steps.get_version.outputs.VERSION }}" >> $GITHUB_STEP_SUMMARY
echo '```' >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "### Published Wheels" >> $GITHUB_STEP_SUMMARY echo "# CUDA version (requires NVIDIA driver with CUDA 11.8+ or 12.x support)" >> $GITHUB_STEP_SUMMARY
echo "Total: $(ls -1 dist/*.whl | wc -l) wheels (3 Python versions: 3.10, 3.11, 3.12)" >> $GITHUB_STEP_SUMMARY echo "pip install kt-kernel-cuda==${{ steps.get_version.outputs.VERSION }}" >> $GITHUB_STEP_SUMMARY
echo '```' >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "### Features" >> $GITHUB_STEP_SUMMARY echo "### Features" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "**CPU wheels with multi-variant support:**" >> $GITHUB_STEP_SUMMARY echo "**kt-kernel (CPU) - Multi-variant support:**" >> $GITHUB_STEP_SUMMARY
echo "- ✅ AMX (Intel Sapphire Rapids+)" >> $GITHUB_STEP_SUMMARY echo "- ✅ AMX (Intel Sapphire Rapids+)" >> $GITHUB_STEP_SUMMARY
echo "- ✅ AVX512 (Intel Skylake-X/Ice Lake/Cascade Lake)" >> $GITHUB_STEP_SUMMARY echo "- ✅ AVX512 (Intel Skylake-X/Ice Lake/Cascade Lake)" >> $GITHUB_STEP_SUMMARY
echo "- ✅ AVX2 (Maximum compatibility)" >> $GITHUB_STEP_SUMMARY echo "- ✅ AVX2 (Maximum compatibility)" >> $GITHUB_STEP_SUMMARY
echo "- 🔧 Runtime CPU detection: Automatically selects optimal variant" >> $GITHUB_STEP_SUMMARY echo "- 🔧 Runtime CPU detection: Automatically selects optimal variant" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "**CUDA wheels with multi-architecture support:**" >> $GITHUB_STEP_SUMMARY echo "**kt-kernel-cuda (CUDA) - Multi-architecture support:**" >> $GITHUB_STEP_SUMMARY
echo "- ✅ SM 80 (Ampere: A100, RTX 3000 series)" >> $GITHUB_STEP_SUMMARY echo "- ✅ SM 80 (Ampere: A100, RTX 3000 series)" >> $GITHUB_STEP_SUMMARY
echo "- ✅ SM 86 (Ampere: RTX 3060-3090)" >> $GITHUB_STEP_SUMMARY echo "- ✅ SM 86 (Ampere: RTX 3060-3090)" >> $GITHUB_STEP_SUMMARY
echo "- ✅ SM 89 (Ada Lovelace: RTX 4000 series)" >> $GITHUB_STEP_SUMMARY echo "- ✅ SM 89 (Ada Lovelace: RTX 4000 series)" >> $GITHUB_STEP_SUMMARY
echo "- ✅ SM 90 (Hopper: H100)" >> $GITHUB_STEP_SUMMARY echo "- ✅ SM 90 (Hopper: H100)" >> $GITHUB_STEP_SUMMARY
echo "- 🔧 Static CUDA runtime: Compatible with CUDA 11.8+ and 12.x drivers" >> $GITHUB_STEP_SUMMARY echo "- 🔧 Static CUDA runtime: Compatible with CUDA 11.8+ and 12.x drivers" >> $GITHUB_STEP_SUMMARY
echo "- 🔧 Includes multi-variant CPU code (AMX/AVX512/AVX2)" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY echo "" >> $GITHUB_STEP_SUMMARY
echo "**Installation:**" >> $GITHUB_STEP_SUMMARY echo "### Links" >> $GITHUB_STEP_SUMMARY
echo '```bash' >> $GITHUB_STEP_SUMMARY echo "- CPU package: https://pypi.org/project/kt-kernel/${{ steps.get_version.outputs.VERSION }}/" >> $GITHUB_STEP_SUMMARY
echo "# CPU version" >> $GITHUB_STEP_SUMMARY echo "- CUDA package: https://pypi.org/project/kt-kernel-cuda/${{ steps.get_version.outputs.VERSION }}/" >> $GITHUB_STEP_SUMMARY
echo "pip install kt-kernel==${{ steps.get_version.outputs.VERSION }}+cpu" >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY
echo "# CUDA version (requires NVIDIA driver with CUDA 11.8+ or 12.x support)" >> $GITHUB_STEP_SUMMARY
echo "pip install kt-kernel==${{ steps.get_version.outputs.VERSION }}+cuda118" >> $GITHUB_STEP_SUMMARY
echo '```' >> $GITHUB_STEP_SUMMARY
echo "" >> $GITHUB_STEP_SUMMARY
echo "PyPI link: https://pypi.org/project/kt-kernel/#history" >> $GITHUB_STEP_SUMMARY

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: Install the latest CPU-only version:
```bash ```bash
pip install "kt-kernel==0.5.0+cpu" pip install kt-kernel
```
Or let pip auto-select the latest CPU version:
```bash
pip install kt-kernel # Defaults to CPU version
``` ```
> **Note**: Check the [latest version on PyPI](https://pypi.org/project/kt-kernel/#history) > **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: For NVIDIA GPU-accelerated inference:
```bash ```bash
pip install "kt-kernel==0.5.0+cuda118" pip install kt-kernel-cuda
``` ```
**Features:** **Features:**

28
kt-kernel/setup.py
View File

@@ -698,31 +698,33 @@ if _version_file.exists():
else: else:
_base_version = "0.5.0" _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: if "CPUINFER_VERSION" in os.environ:
# User explicitly set version (e.g., for testing) # User explicitly set version (e.g., for testing)
VERSION = os.environ["CPUINFER_VERSION"] VERSION = os.environ["CPUINFER_VERSION"]
print(f"-- Explicit version: {VERSION}") print(f"-- Explicit version: {VERSION}")
else: else:
# Auto-detect suffix based on CUDA usage VERSION = _base_version
cuda_enabled = _env_get_bool("CPUINFER_USE_CUDA", False)
if cuda_enabled: # Determine package name based on CUDA usage
# CUDA build: add +cuda118 suffix cuda_enabled = _env_get_bool("CPUINFER_USE_CUDA", False)
# (CUDA 11.8 is the build toolkit version for compatibility with 11.8+ and 12.x) if cuda_enabled:
VERSION = f"{_base_version}+cuda118" # CUDA build: use kt-kernel-cuda package name
print(f"-- CUDA wheel version: {VERSION}") # Compatible with CUDA 11.8+ and 12.x drivers
else: PACKAGE_NAME = "kt-kernel-cuda"
# CPU-only build: add +cpu suffix print(f"-- CUDA wheel: {PACKAGE_NAME} version {VERSION}")
VERSION = f"{_base_version}+cpu" else:
print(f"-- CPU wheel version: {VERSION}") # CPU-only build: use kt-kernel package name
PACKAGE_NAME = "kt-kernel"
print(f"-- CPU wheel: {PACKAGE_NAME} version {VERSION}")
################################################################################ ################################################################################
# Setup # Setup
################################################################################ ################################################################################
setup( setup(
name="kt-kernel", name=PACKAGE_NAME,
version=VERSION, version=VERSION,
description="KT-Kernel: High-performance kernel operations for KTransformers (AMX/AVX/KML optimizations)", description="KT-Kernel: High-performance kernel operations for KTransformers (AMX/AVX/KML optimizations)",
author="kvcache-ai", author="kvcache-ai",