Windows AI Wheelhouse

面向 Windows 平台的 AI 预编译 Python 轮子合集

⏱️ 最后自动更新（UTC）：2025-11-29T21:15:19.614430+00:00

📑 目录

项目愿景
致敬与启发
快速开始
- 环境准备
- 安装示例
CXX11 ABI 说明
可用轮子
贡献
致谢

🟦 项目愿景

本项目作为一个学习和实践的项目，在延续 wildminder AI-windows-whl 优秀理念的基础上，尝试通过以下方式提供一些补充：

🤖 自动化更新：通过 GitHub Actions 定期自动抓取和更新 wheel 链接
📊 多源聚合：整合来自不同社区贡献者的预编译资源
📖 结构化呈现：按版本和依赖关系分类展示，便于查找
🔍 智能解析：自动识别版本号、CXX11 ABI 等元数据

这些尝试完全出于个人学习目的，也希望能够为社区提供一个不同角度的资讯汇整。如果本项目能为更多 Windows AI 爱好者节省一点点时间，让大家能够更专注于创作和探索，那便这个项目最大的荣幸。

🟦 致敬与启发

本项目深受 wildminder AI-windows-whl 项目的启发，我们旨在延续其精神，为更广泛的 Windows AI 社区提供稳定可靠的预编译资源。

特别感谢 Eddy 孜孜不倦的精神和无私的指导。从接触他的 Palingenesis 模型开始，是他的热情感召，让我这样一个影视从业者，能够开始尝试撰写自己需要的代码。

我的 Palingenesis AIGC 作品

使用 Eddy 的 Palingenesis 模型创作的 MV：

点击图片观看视频

🟦 快速开始

按照以下步骤即可使用仓库中的 wheel：

环境准备

Windows 版 Python：请确认本地安装的 Python 版本与 PyTorch 支持范围一致（当前为 3.9 - 3.14）。可前往 Python 官网下载。

安装示例

通过 pip 搭配 .whl 直链进行安装，建议给 URL 加上引号，防止 shell 解析问题。

# 以安装特定 flash-attention 轮子为例
pip install "https://huggingface.co/lldacing/flash-attention-windows-wheel/blob/main/flash_attn-2.7.4.post1+cu128torch2.7.0cxx11abiFALSE-cp312-cp312-win_amd64.whl"

在下方可用轮子区域按 Python/PyTorch/CUDA 版本找到匹配行，然后复制下载链接放入 pip install 命令即可。

(返回顶部)

🟦 CXX11 ABI 说明

部分表格包含 CXX11 ABI 列，这是 C++ 编译器的两种不兼容模式。表格中使用以下符号标注：

✓ 或 **TRUE**：启用 CXX11 ABI（新版，适用于新版 PyTorch，通常 2.7.0 及以上）
✗ 或 **FALSE**：禁用 CXX11 ABI（Pre-CXX11 ABI，旧版）
**—**（横线或未标注）：未明确标注（大多数 Windows 用户应优先选择此版本）

PyTorch 和扩展库的 CXX11 ABI 必须一致，否则会导致崩溃或链接错误。

如何检测你的 PyTorch 使用哪种 ABI？

运行以下 Python 代码：

import torch
print(f"PyTorch 版本: {torch.__version__}")
print(f"CXX11 ABI: {torch._C._GLIBCXX_USE_CXX11_ABI}")

结果解读：

输出 0 或 False → 使用 Pre-CXX11 ABI，选择 CXX11 ABI 列为 —、✗ 或 FALSE 的 wheel
输出 1 或 True → 使用 CXX11 ABI，选择 CXX11 ABI 列为 ✓ 或 TRUE 的 wheel

快速选择指南

对于大多数 Windows 用户（使用 PyTorch 官方版本）：

✅ 优先选择 CXX11 ABI 列显示为 — 的 wheel（最兼容）
✅ 次选 CXX11 ABI 列显示为 ✗ 或 FALSE 的 wheel（明确兼容）
⚠️ 仅在确认后选择 CXX11 ABI 列显示为 ✓ 或 TRUE 的 wheel

如果不确定，优先选择未标注 CXX11 ABI（显示 —）的版本，这些版本通常与官方 PyTorch 兼容。

(返回顶部)

🟦 可用轮子

以下是当前收录的轮子

⚙️ PyTorch

深度学习框架的基石，强烈建议优先通过官方渠道安装以确保兼容性。

官方安装页面： https://pytorch.org/get-started/locally/

为了方便，在 Linux/WSL（NVIDIA GPU）环境下列出常用命令；其他配置（CPU、macOS、ROCm）请参考官网指引。

稳定版（2.9.1）

多数用户推荐使用该版本

CUDA Version	Pip Install Command
CUDA 13.0	`pip install torch torchvision --index-url https://download.pytorch.org/whl/cu130`
CUDA 12.8	`pip install torch torchvision --index-url https://download.pytorch.org/whl/cu128`
CUDA 12.6	`pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126`

历史稳定版 2.7.1 / 2.8.0

历史稳定版（2.8.0）

CUDA Version	Pip Install Command
CUDA 12.9	`pip install "torch>=2.8.0.dev,<2.9.0" torchvision --index-url https://download.pytorch.org/whl/cu129`
CUDA 12.8	`pip install "torch>=2.8.0.dev,<2.9.0" torchvision --index-url https://download.pytorch.org/whl/cu128`
CUDA 12.6	`pip install "torch>=2.8.0.dev,<2.9.0" torchvision --index-url https://download.pytorch.org/whl/cu126`

历史稳定版（2.7.1）

CUDA Version	Pip Install Command
CUDA 12.8	`pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu128`
CUDA 12.6	`pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu126`
CUDA 11.8	`pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cu118`
CPU only	`pip install torch==2.7.1 torchvision==0.22.1 torchaudio==2.7.1 --index-url https://download.pytorch.org/whl/cpu`

Nightly 版本

适合追新特性，但可能存在不稳定。

PyTorch 2.10（Nightly）

CUDA Version	Pip Install Command
CUDA 13.0	`pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu130`
CUDA 12.8	`pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128`
CUDA 12.6	`pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu126`

⚙️ Torchaudio

PyTorch 官方音频处理库，提供音频 I/O、转换和特征提取等功能。

官方仓库： https://github.com/pytorch/audio
预编译来源： Wildminder

展开已收录的 Torchaudio

Package Version	Python Ver	PyTorch Ver	CUDA Ver	CXX11 ABI	Source	Download Link
`2.8.0a0`	`3.12`	`2.9.0`	`12.8`	✓	Wildminder	Link
`2.8.0a0`	`3.13`	`2.10.0`	`13.0`	✓	Wildminder	Link

(返回顶部)

⚙️ Flash Attention

高性能注意力机制的先驱实现，通过分块计算和重计算策略大幅降低显存并提速。

官方仓库： Dao-AILab/flash-attention
预编译来源： Wildminder、lldacing、mjun0812、BradPita、Kijai

展开已收录的 Flash Attention

📝 CXX11 ABI 说明：

✓ = 启用 CXX11 ABI（适用于新版 PyTorch，通常 2.7.0 及以上）

✗ = 禁用 CXX11 ABI（Pre-CXX11 ABI，适用于旧版或特定配置）

— = 未明确标注（请根据您的 PyTorch 版本选择，或参考 CXX11 ABI 说明）

Package Version	Python Ver	PyTorch Ver	CUDA Ver	CXX11 ABI	Source	Download Link
`2.5.9`	`3.10`	`2.4`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.4`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.5`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.5`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.6`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.6`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.7`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.10`	`2.7`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.4`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.4`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.5`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.5`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.6`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.6`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.7`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.11`	`2.7`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.4`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.4`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.5`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.5`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.6`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.6`	`12.8`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.7`	`12.4`	—	mjun0812	Link
`2.5.9`	`3.12`	`2.7`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.4`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.4`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.5`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.5`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.6`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.6`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.7`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.10`	`2.7`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.4`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.4`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.5`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.5`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.6`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.6`	`12.6`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.6`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.7`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.11`	`2.7`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.4`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.4`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.5`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.5`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.6`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.6`	`12.8`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.7`	`12.4`	—	mjun0812	Link
`2.6.3`	`3.12`	`2.7`	`12.8`	—	mjun0812	Link
`2.7.0.post2`	`3.10`	`2.4.0`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.10`	`2.4.1`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.11`	`2.4.0`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.11`	`2.4.1`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.11`	`2.5.0`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.11`	`2.5.1`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.12`	`2.4.0`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.12`	`2.4.1`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.12`	`2.5.0`	`12.4`	✗	lldacing	Link
`2.7.0.post2`	`3.12`	`2.5.1`	`12.4`	✗	lldacing	Link
`2.7.4`	`3.10`	`2.4`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.10`	`2.5`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.5`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.10`	`2.5.1`	`12.4`	✗	lldacing	Link
`2.7.4`	`3.10`	`2.6.0`	`12.4`	✗	lldacing	Link
`2.7.4`	`3.10`	`2.6`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.6`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.10`	`2.6.0`	`12.6`	✗	lldacing	Link
`2.7.4.post1`	`3.10`	`2.6`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.10`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.7.0`	`12.8`	✗	lldacing	Link
`2.7.4`	`3.10`	`2.7`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.8`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.8.0`	`12.8`	✓	lldacing	Link
`2.7.4`	`3.10`	`2.8`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.9`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.10`	`2.9`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.11`	`2.5`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.5`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.11`	`2.6.0`	`12.4`	✗	lldacing	Link
`2.7.4.post1`	`3.11`	`2.6`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.11`	`2.6.0`	`12.6`	✗	lldacing	Link
`2.7.4.post1`	`3.11`	`2.6`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.11`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.7`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.7.0`	`12.8`	✗	lldacing	Link
`2.7.4`	`3.11`	`2.7`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.8`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.11`	`2.8`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.11`	`2.9`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.12`	`2.4`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.12`	`2.5`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.12`	`2.6.0`	`12.4`	✗	lldacing	Link
`2.7.4`	`3.12`	`2.6`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.6`	`12.4`	—	mjun0812	Link
`2.7.4`	`3.12`	`2.6.0`	`12.6`	✗	lldacing	Link
`2.7.4.post1`	`3.12`	`2.6`	`12.6`	—	mjun0812	Link
`2.7.4`	`3.12`	`2.6.0`	`12.8`	✗	Kijai	Link
`2.7.4`	`3.12`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.7`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.7.0`	`12.8`	✗	lldacing	Link
`2.7.4`	`3.12`	`2.7`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.8`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.8.0`	`12.8`	✓	Wildminder	Link
`2.7.4`	`3.12`	`2.8`	`12.8`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.9`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.12`	`2.9`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.6`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.7`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.7`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.8`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.8`	`12.6`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.9`	`12.4`	—	mjun0812	Link
`2.7.4.post1`	`3.13`	`2.9`	`12.6`	—	mjun0812	Link
`2.8.0.post2`	`3.12`	`2.8.0`	`12.8`	✓	Wildminder	Link
`2.8.1`	`3.12`	`2.8.0`	`12.8`	✓	Wildminder	Link
`2.8.2`	`3.10`	`2.7`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.10`	`2.8`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.11`	`2.7`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.11`	`2.8`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.12`	`2.7`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.12`	`2.8.0`	`12.8`	✓	Wildminder	Link
`2.8.2`	`3.12`	`2.8`	`12.8`	—	mjun0812	Link
`2.8.2`	`3.12`	`2.9.0`	`12.8`	✓	Wildminder	Link
`2.8.2`	`3.13`	`2.6`	`12.4`	—	mjun0812	Link
`2.8.2`	`3.13`	`2.7`	`12.4`	—	mjun0812	Link
`2.8.2`	`3.13`	`2.7`	`12.6`	—	mjun0812	Link
`2.8.2`	`3.13`	`2.8`	`12.4`	—	mjun0812	Link
`2.8.2`	`3.13`	`2.8`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.5`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.5`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.6`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.6`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.7`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.7`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.8`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.8`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.9`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.10`	`2.9`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.5`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.5`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.6`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.6`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.7`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.7`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.8`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.8`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.9`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.9`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.11`	`2.9`	`13.0`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.5`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.5`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.6`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.6`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.7`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.7`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.8`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.8`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.8.0`	`12.8`	✓	Wildminder	Link
`2.8.3`	`3.12`	`2.9`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.9`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.12`	`2.9.0`	`12.8`	✓	Wildminder	Link
`2.8.3`	`3.12`	`2.9.0`	`13.0`	✓	Wildminder	Link
`2.8.3`	`3.12`	`2.9.1`	`13.0`	✓	Wildminder	Link
`2.8.3`	`3.12`	`2.10.0`	`13.0`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.6`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.6`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.7`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.7`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.8`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.8`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.8.0`	`12.8`	✓	BradPita	Link
`2.8.3`	`3.13`	`2.9`	`12.4`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.9`	`12.6`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.9.0`	`12.9`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.9.0`	`13.0`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.9`	`13.0`	—	mjun0812	Link
`2.8.3`	`3.13`	`2.9.1`	`12.8`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.9.1`	`13.0`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.10.0`	`12.8`	✓	Wildminder	Link
`2.8.3`	`3.13`	`2.10.0`	`13.0`	✓	Wildminder	Link

(返回顶部)

⚙️ xformers

Meta 开发的内存高效注意力库，提供多种优化算子，广泛应用于图像生成和 LLM 推理。

官方仓库： facebookresearch/xformers
预编译来源： PyTorch、Wildminder

安装 PyTorch 后通常可以直接 pip install xformers。若失败，可在预编译来源中寻找与环境匹配的 wheel。

ABI3 版本可兼容 Python 3.9-3.12。

展开已收录的 xformers

Package Version	Python Ver	PyTorch Ver	CUDA Ver	Source	Download Link
`0.0.22.post2`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post3`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post7`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post7`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post2`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post3`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post7`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post7`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post2`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post3`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post7`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post7`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post2`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post3`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post7`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.22.post4`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.22.post7`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23.post1`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23.post1`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23.post1`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23.post1`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23.post1`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23.post1`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23.post1`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.23`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.23.post1`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.24`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.24`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.24`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.24`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.24`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.24`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.24`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.24`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25.post1`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25.post1`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25.post1`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25.post1`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25.post1`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25.post1`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25.post1`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.25`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.25.post1`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26.post1`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26.post1`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26.post1`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26.post1`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26.post1`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26.post1`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26.post1`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.26`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.26.post1`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post1`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post2`	`3.8`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post1`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post2`	`3.8`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post1`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post2`	`3.9`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post1`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post2`	`3.9`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post1`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post2`	`3.10`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post1`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post2`	`3.10`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post1`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post2`	`3.11`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post1`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post2`	`3.11`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27`	`3.12`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post1`	`3.12`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27.post2`	`3.12`	`2.0-2.4`	`11.8`	PyTorch	Link
`0.0.27`	`3.12`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post1`	`3.12`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.27.post2`	`3.12`	`2.1-2.5`	`12.1`	PyTorch	Link
`0.0.28.post1`	`3.8`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post1`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post2`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post3`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post1`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post2`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post3`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post1`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post2`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post3`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post1`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post2`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.28.post3`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post1`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post2`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.9`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.9`	`2.5-2.6`	`12.6`	PyTorch	Link
`0.0.29`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post1`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post2`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.10`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.10`	`2.5-2.6`	`12.6`	PyTorch	Link
`0.0.29`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post1`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post2`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.11`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.11`	`2.5-2.6`	`12.6`	PyTorch	Link
`0.0.29`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post1`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post2`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.12`	`2.4-2.6`	`12.4`	PyTorch	Link
`0.0.29.post3`	`3.12`	`2.5-2.6`	`12.6`	PyTorch	Link
`0.0.30`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.30`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.30`	`3.10`	`2.5+`	`12.6`	PyTorch	Link
`0.0.30`	`3.10`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.30`	`3.11`	`2.5+`	`12.6`	PyTorch	Link
`0.0.30`	`3.11`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.30`	`3.12`	`2.5+`	`12.6`	PyTorch	Link
`0.0.30`	`3.12`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.31`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.31.post1`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.31`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.31.post1`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.32.post1`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.32.post2`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.32.post1`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.32.post2`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.32.post1`	`3.9`	`2.8-2.9`	`12.9`	PyTorch	Link
`0.0.32.post2`	`3.9`	`2.8-2.9`	`12.9`	PyTorch	Link
`0.0.33`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.33.post1`	`3.9`	`2.5+`	`12.6`	PyTorch	Link
`0.0.33`	`3.9`	`2.9+`	`13.0`	PyTorch	Link
`0.0.33.post1`	`3.9`	`2.9+`	`13.0`	PyTorch	Link
`0.0.33`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.33.post1`	`3.9`	`2.7-2.9`	`12.8`	PyTorch	Link
`0.0.33`	`3.9`	`2.9`	`13.0`	Wildminder	Link
`0.0.33`	`3.9`	`2.10`	`13.0`	Wildminder	Link

(返回顶部)

⚙️ SageAttention

精准且高效的注意力机制，通过平滑量化大幅降低显存占用，同时保持精度。

官方仓库： thu-ml/SageAttention
预编译来源： woct0rdho、sdbds

展开已收录的 SageAttention

Package Version	Python Ver	PyTorch Ver	CUDA Ver	Source	Download Link
`2.0.1`	`3.10`	`2.5.1`	`12.4`	sdbds	Link
`2.1.1`	`3.9`	`2.5.1`	`12.4`	woct0rdho	Link
`2.1.1`	`3.9`	`2.6.0`	`12.6`	woct0rdho	Link
`2.1.1`	`3.9`	`2.7.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.9`	`2.7.1`	`12.8`	woct0rdho	Link
`2.1.1`	`3.9`	`2.8.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.10`	`2.5.1`	`12.4`	woct0rdho	Link
`2.1.1`	`3.10`	`2.6.0`	`12.6`	woct0rdho	Link
`2.1.1`	`3.10`	`2.7.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.10`	`2.7.1`	`12.8`	woct0rdho	Link
`2.1.1`	`3.10`	`2.8.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.11`	`2.5.1`	`12.4`	woct0rdho	Link
`2.1.1`	`3.11`	`2.6.0`	`12.6`	woct0rdho	Link
`2.1.1`	`3.11`	`2.7.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.11`	`2.7.1`	`12.8`	woct0rdho	Link
`2.1.1`	`3.11`	`2.8.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.12`	`2.5.1`	`12.4`	woct0rdho	Link
`2.1.1`	`3.12`	`2.6.0`	`12.6`	woct0rdho	Link
`2.1.1`	`3.12`	`2.7.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.12`	`2.7.1`	`12.8`	woct0rdho	Link
`2.1.1`	`3.12`	`2.8.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.13`	`2.6.0`	`12.6`	woct0rdho	Link
`2.1.1`	`3.13`	`2.7.0`	`12.8`	woct0rdho	Link
`2.1.1`	`3.13`	`2.7.1`	`12.8`	woct0rdho	Link
`2.1.1`	`3.13`	`2.8.0`	`12.8`	woct0rdho	Link

(返回顶部)

⚙️ SageAttention 2.2 (SageAttention2++)

SageAttention 的升级版本，引入更先进的量化策略和优化算法，显存节省更多，速度更快。

官方仓库： thu-ml/SageAttention
预编译来源： woct0rdho、sdbds、Wildminder、Eddy

仅支持 CUDA ≥ 12.8，因此需搭配 PyTorch ≥ 2.7。

展开已收录的 SageAttention 2++

Package Version	Python Ver	PyTorch Ver	CUDA Ver	Source	Download Link
`2.2.0.post1`	`3.9`	`2.5.1`	`12.4`	woct0rdho	Link
`2.2.0.post2`	`3.9`	`2.5.1`	`12.4`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.5.1`	`12.4`	woct0rdho	Link
`2.2.0.post1`	`3.9`	`2.6.0`	`12.6`	woct0rdho	Link
`2.2.0.post2`	`3.9`	`2.6.0`	`12.6`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.6.0`	`12.6`	woct0rdho	Link
`2.2.0.post1`	`3.9`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0.post2`	`3.9`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0`	`3.9`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0.post1`	`3.9`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0.post2`	`3.9`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0`	`3.9`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.9.0`	`12.8`	woct0rdho	Link
`2.2.0.post4`	`3.9`	`2.9.0`	`12.8`	woct0rdho	Link
`2.2.0.post3`	`3.9`	`2.9.0`	`13.0`	woct0rdho	Link
`2.2.0.post4`	`3.9`	`2.9.0`	`13.0`	woct0rdho	Link
`2.2.0`	`3.10`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0`	`3.10`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0`	`3.11`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0`	`3.11`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0`	`3.11`	`2.9.0`	`13.0`	sdbds	Link
`2.2.0`	`3.11`	`2.9.1`	`13.0`	sdbds	Link
`2.2.0`	`3.12`	`2.7.0`	`12.8`	Eddy	Link
`2.2.0`	`3.12`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0`	`3.12`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0`	`3.12`	`2.9.0`	`12.8`	Wildminder	Link
`2.2.0.post3`	`3.12`	`2.10.0`	`12.8`	Wildminder	Link
`2.2.0.post3`	`3.12`	`2.10.0`	`13.0`	Wildminder	Link
`2.2.0`	`3.13`	`2.7.1`	`12.8`	woct0rdho	Link
`2.2.0`	`3.13`	`2.8.0`	`12.8`	woct0rdho	Link
`2.2.0.post3`	`3.13`	`2.8.0`	`12.8`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.8.0`	`12.9`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.9.0`	`12.8`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.9.0`	`12.9`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.9.0`	`13.0`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.10.0`	`12.8`	Wildminder	Link
`2.2.0.post3`	`3.13`	`2.10.0`	`13.0`	Wildminder	Link

(返回顶部)

⚙️ SageAttention 3

用于推理的微缩放 FP4 注意力机制和 8 位训练的探索，目前仅支持 Blackwell 架构。

官方仓库： thu-ml/SageAttention
预编译来源： sdbds、Eddy

SageAttention3 不保證所有型號的無損加速。對於其他影片產生模型,我們建議在某些圖層或時間步長中選擇性地使用 SageAttention2++

展开已收录的 SageAttention 3

Package Version	Python Ver	PyTorch Ver	CUDA Ver	Source	Download Link
`1.0.0`	`3.11`	`2.8.0`	`12.8`	sdbds	Link
`1.0.0`	`3.12`	`2.8.0`	`12.8`	Eddy	Link

(返回顶部)

⚙️ SpargeAttn

稀疏注意力机制，通过智能跳过不重要的计算来加速推理，适用于长上下文场景。

官方仓库： thu-ml/SpargeAttn
预编译来源： woct0rdho

展开已收录的 SpargeAttn

Package Version	PyTorch Ver	CUDA Ver	Source	Download Link
`0.1.0.post2`	`2.5.1`	`12.4`	woct0rdho	Link
`0.1.0.post3`	`2.5.1`	`12.4`	woct0rdho	Link
`0.1.0`	`2.5.1`	`12.4`	woct0rdho	Link
`0.1.0.post2`	`2.6.0`	`12.6`	woct0rdho	Link
`0.1.0.post3`	`2.6.0`	`12.6`	woct0rdho	Link
`0.1.0`	`2.6.0`	`12.6`	woct0rdho	Link
`0.1.0.post1`	`2.7.1`	`12.8`	woct0rdho	Link
`0.1.0.post2`	`2.7.1`	`12.8`	woct0rdho	Link
`0.1.0.post3`	`2.7.1`	`12.8`	woct0rdho	Link
`0.1.0`	`2.7.1`	`12.8`	woct0rdho	Link
`0.1.0.post1`	`2.8.0`	`12.8`	woct0rdho	Link
`0.1.0.post2`	`2.8.0`	`12.8`	woct0rdho	Link
`0.1.0.post3`	`2.8.0`	`12.8`	woct0rdho	Link
`0.1.0`	`2.8.0`	`12.8`	woct0rdho	Link
`0.1.0.post2`	`2.9.0`	`12.8`	woct0rdho	Link
`0.1.0.post3`	`2.9.0`	`12.8`	woct0rdho	Link
`0.1.0.post3`	`2.9.0`	`13.0`	woct0rdho	Link

(返回顶部)

⚙️ Nunchaku

MIT 开发的 Transformer 推理加速库，专注于提升解码速度和吞吐量。

官方仓库： mit-han-lab/nunchaku
预编译来源： nunchaku-tech、Wildminder

展开已收录的 Nunchaku

Package Version	Python Ver	PyTorch Ver	Source	Download Link
`0.2.0`	`3.10`	`2.5`	nunchaku-tech	Link
`0.2.0`	`3.10`	`2.6`	nunchaku-tech	Link
`0.2.0`	`3.10`	`2.7`	nunchaku-tech	Link
`0.2.0`	`3.10`	`2.8`	nunchaku-tech	Link
`0.2.0`	`3.11`	`2.5`	nunchaku-tech	Link
`0.2.0`	`3.11`	`2.6`	nunchaku-tech	Link
`0.2.0`	`3.11`	`2.7`	nunchaku-tech	Link
`0.2.0`	`3.11`	`2.8`	nunchaku-tech	Link
`0.2.0`	`3.12`	`2.5`	nunchaku-tech	Link
`0.2.0`	`3.12`	`2.6`	nunchaku-tech	Link
`0.2.0`	`3.12`	`2.7`	nunchaku-tech	Link
`0.2.0`	`3.12`	`2.8`	nunchaku-tech	Link
`0.2.0`	`3.13`	`2.6`	nunchaku-tech	Link
`0.2.0`	`3.13`	`2.7`	nunchaku-tech	Link
`0.2.0`	`3.13`	`2.8`	nunchaku-tech	Link
`0.3.0`	`3.10`	`2.5`	nunchaku-tech	Link
`0.3.0`	`3.10`	`2.6`	nunchaku-tech	Link
`0.3.0`	`3.10`	`2.7`	nunchaku-tech	Link
`0.3.0`	`3.10`	`2.8`	nunchaku-tech	Link
`0.3.0`	`3.11`	`2.5`	nunchaku-tech	Link
`0.3.0`	`3.11`	`2.6`	nunchaku-tech	Link
`0.3.0`	`3.11`	`2.7`	nunchaku-tech	Link
`0.3.0`	`3.11`	`2.8`	nunchaku-tech	Link
`0.3.0`	`3.12`	`2.5`	nunchaku-tech	Link
`0.3.0`	`3.12`	`2.6`	nunchaku-tech	Link
`0.3.0`	`3.12`	`2.7`	nunchaku-tech	Link
`0.3.0`	`3.12`	`2.8`	nunchaku-tech	Link
`0.3.1`	`3.10`	`2.5`	nunchaku-tech	Link
`0.3.1`	`3.10`	`2.6`	nunchaku-tech	Link
`0.3.1`	`3.10`	`2.7`	nunchaku-tech	Link
`0.3.1`	`3.10`	`2.8`	nunchaku-tech	Link
`0.3.1`	`3.11`	`2.5`	nunchaku-tech	Link
`0.3.1`	`3.11`	`2.6`	nunchaku-tech	Link
`0.3.1`	`3.11`	`2.7`	nunchaku-tech	Link
`0.3.1`	`3.11`	`2.8`	nunchaku-tech	Link
`0.3.1`	`3.12`	`2.5`	nunchaku-tech	Link
`0.3.1`	`3.12`	`2.6`	nunchaku-tech	Link
`0.3.1`	`3.12`	`2.7`	nunchaku-tech	Link
`0.3.1`	`3.12`	`2.8`	nunchaku-tech	Link
`0.3.2`	`3.10`	`2.5`	nunchaku-tech	Link
`0.3.2`	`3.10`	`2.6`	nunchaku-tech	Link
`0.3.2`	`3.10`	`2.7`	nunchaku-tech	Link
`0.3.2`	`3.10`	`2.8`	nunchaku-tech	Link
`0.3.2`	`3.10`	`2.9`	nunchaku-tech	Link
`0.3.2`	`3.11`	`2.5`	nunchaku-tech	Link
`0.3.2`	`3.11`	`2.6`	nunchaku-tech	Link
`0.3.2`	`3.11`	`2.7`	nunchaku-tech	Link
`0.3.2`	`3.11`	`2.8`	nunchaku-tech	Link
`0.3.2`	`3.11`	`2.9`	nunchaku-tech	Link
`0.3.2`	`3.12`	`2.5`	nunchaku-tech	Link
`0.3.2`	`3.12`	`2.6`	nunchaku-tech	Link
`0.3.2`	`3.12`	`2.7`	nunchaku-tech	Link
`0.3.2`	`3.12`	`2.8`	nunchaku-tech	Link
`0.3.2`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.0`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250820`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.0`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.0`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.0`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250820`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.0`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250820`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.0`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.0`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.0`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.0`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.0`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.0`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.0`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.0`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.0`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250823`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250830`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250902`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.0`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.0`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.0`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.0`	`3.13`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250903`	`3.13`	`2.9`	nunchaku-tech	Link
`1.0.0.dev20250904`	`3.13`	`2.9`	nunchaku-tech	Link
`1.0.1`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.10`	`2.5`	nunchaku-tech	Link
`1.0.1`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.10`	`2.6`	nunchaku-tech	Link
`1.0.1`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.1`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.1`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.1`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.11`	`2.5`	nunchaku-tech	Link
`1.0.1`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.11`	`2.6`	nunchaku-tech	Link
`1.0.1`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.1`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.1`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.1`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.12`	`2.5`	nunchaku-tech	Link
`1.0.1`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.12`	`2.6`	nunchaku-tech	Link
`1.0.1`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.1`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.1`	`3.12`	`2.9`	Wildminder	Link
`1.0.1`	`3.12`	`2.9`	Wildminder	Link
`1.0.1`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.1`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.13`	`2.6`	nunchaku-tech	Link
`1.0.1`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.1`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.1`	`3.13`	`2.9`	Wildminder	Link
`1.0.1`	`3.13`	`2.9`	Wildminder	Link
`1.0.1`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250923`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250924`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250926`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250929`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.1.dev20250930`	`3.13`	`2.10`	nunchaku-tech	Link
`1.0.2`	`3.10`	`2.7`	nunchaku-tech	Link
`1.0.2`	`3.10`	`2.8`	nunchaku-tech	Link
`1.0.2`	`3.10`	`2.9`	nunchaku-tech	Link
`1.0.2`	`3.10`	`2.10`	nunchaku-tech	Link
`1.0.2`	`3.11`	`2.7`	nunchaku-tech	Link
`1.0.2`	`3.11`	`2.8`	nunchaku-tech	Link
`1.0.2`	`3.11`	`2.9`	nunchaku-tech	Link
`1.0.2`	`3.11`	`2.10`	nunchaku-tech	Link
`1.0.2`	`3.12`	`2.7`	nunchaku-tech	Link
`1.0.2`	`3.12`	`2.8`	nunchaku-tech	Link
`1.0.2`	`3.12`	`2.9`	nunchaku-tech	Link
`1.0.2`	`3.12`	`2.10`	nunchaku-tech	Link
`1.0.2`	`3.13`	`2.7`	nunchaku-tech	Link
`1.0.2`	`3.13`	`2.8`	nunchaku-tech	Link
`1.0.2`	`3.13`	`2.9`	nunchaku-tech	Link
`1.0.2`	`3.13`	`2.10`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.10`	`2.9`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.10`	`2.10`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.11`	`2.9`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.11`	`2.10`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.12`	`2.9`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.12`	`2.10`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.13`	`2.9`	nunchaku-tech	Link
`1.1.0.dev20251111`	`3.13`	`2.10`	nunchaku-tech	Link

(返回顶部)

⚙️ NATTEN (Neighborhood Attention Transformer)

局部邻域注意力实现，在视觉任务中表现出色。

官方仓库： SHI-Labs/NATTEN
预编译来源： lldacing

展开已收录的 NATTEN

Package Version	Python Ver	PyTorch Ver	CUDA Ver	Source	Download Link
`0.17.3`	`3.10`	`2.4.0`	`12.4`	lldacing	Link
`0.17.3`	`3.10`	`2.4.1`	`12.4`	lldacing	Link
`0.17.3`	`3.10`	`2.5.0`	`12.4`	lldacing	Link
`0.17.3`	`3.10`	`2.5.1`	`12.4`	lldacing	Link
`0.17.3`	`3.11`	`2.4.0`	`12.4`	lldacing	Link
`0.17.3`	`3.11`	`2.4.1`	`12.4`	lldacing	Link
`0.17.3`	`3.11`	`2.5.0`	`12.4`	lldacing	Link
`0.17.3`	`3.11`	`2.5.1`	`12.4`	lldacing	Link
`0.17.3`	`3.12`	`2.4.0`	`12.4`	lldacing	Link
`0.17.3`	`3.12`	`2.4.1`	`12.4`	lldacing	Link
`0.17.3`	`3.12`	`2.5.0`	`12.4`	lldacing	Link
`0.17.3`	`3.12`	`2.5.1`	`12.4`	lldacing	Link
`0.17.5`	`3.10`	`2.6.0`	`12.6`	lldacing	Link
`0.17.5`	`3.10`	`2.7.0`	`12.8`	lldacing	Link
`0.17.5`	`3.11`	`2.6.0`	`12.6`	lldacing	Link
`0.17.5`	`3.11`	`2.7.0`	`12.8`	lldacing	Link
`0.17.5`	`3.12`	`2.6.0`	`12.6`	lldacing	Link
`0.17.5`	`3.12`	`2.7.0`	`12.8`	lldacing	Link

(返回顶部)

⚙️ Triton（Windows Fork）

用于编写高效深度学习算子的语言与编译器，官方不支持 Windows，这是社区维护的 Fork 版本。

Windows Fork： woct0rdho
安装命令： pip install -U "triton-windows<3.6"

展开已收录的 Triton

Package Version	Python Ver	Source	Download Link
`3.0.0`	`3.8`	woct0rdho	Link
`3.0.0`	`3.9`	woct0rdho	Link
`3.0.0`	`3.10`	woct0rdho	Link
`3.0.0`	`3.11`	woct0rdho	Link
`3.0.0`	`3.12`	woct0rdho	Link
`3.1.0`	`3.8`	woct0rdho	Link
`3.1.0`	`3.9`	woct0rdho	Link
`3.1.0`	`3.10`	woct0rdho	Link
`3.1.0`	`3.11`	woct0rdho	Link
`3.1.0`	`3.12`	woct0rdho	Link
`3.2.0`	`3.9`	woct0rdho	Link
`3.2.0`	`3.10`	woct0rdho	Link
`3.2.0`	`3.11`	woct0rdho	Link
`3.2.0`	`3.12`	woct0rdho	Link
`3.2.0`	`3.13`	woct0rdho	Link

(返回顶部)

⚙️ bitsandbytes

轻量级 CUDA 算子库，提供 8-bit 优化器和 LLM 量化支持，大幅降低显存需求。

官方仓库： bitsandbytes-foundation/bitsandbytes
预编译来源： jllllll、bitsandbytes-foundation

展开已收录的 bitsandbytes

Package Version	Source	Download Link
`0.35.0`	jllllll	Link
`0.35.4`	jllllll	Link
`0.37.2`	jllllll	Link
`0.38.1`	jllllll	Link
`0.39.0`	jllllll	Link
`0.39.1`	jllllll	Link
`0.40.0`	jllllll	Link
`0.40.0.post4`	jllllll	Link
`0.40.1.post1`	jllllll	Link
`0.40.2`	jllllll	Link
`0.41.0`	jllllll	Link
`0.41.1`	jllllll	Link
`0.41.2.post2`	jllllll	Link
`0.43.0`	bitsandbytes-foundation	Link
`0.43.1`	bitsandbytes-foundation	Link
`0.43.2`	bitsandbytes-foundation	Link
`1.0.0`	bitsandbytes-foundation	Link
`1.33.7.preview`	bitsandbytes-foundation	Link

(返回顶部)

⚙️ RadialAttention（ComfyUI 节点）

用于长视频生成的径向注意力机制，在保持质量的同时大幅减少内存占用。

官方仓库： mit-han-lab/radial-attention
ComfyUI 节点： ComfyUI-RadialAttn

💡 安装说明：
RadialAttention 不是独立的 wheel 包，而是通过 ComfyUI 自定义节点使用。安装步骤：

先安装 SpargeAttn（见上方，我们已收录其 Windows wheels）

将 ComfyUI-RadialAttn clone 到 ComfyUI/custom_nodes/ 目录

（可选）安装 SageAttention 并使用 --use-sage-attention 参数启动 ComfyUI，作为备用加速方案

(返回顶部)

🟦 贡献 (Contributing)

贡献让开源社区成为一个令人惊叹的学习、启发和创造的地方。我们非常感激您做出的任何贡献。

如果您发现了新的预编译 wheel 文件或可靠的来源，请 pull request，或者直接开一个 issue 附上链接。

🟦 致谢 (Acknowledgments)

本仓库仅仅是一个链接集合。衷心感谢那些为社区辛勤构建和托管这些 wheel 文件的个人和团队。

(返回顶部)

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support