AI搜索登录

开源项目

thu-ml/SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

更新:2026/10/12 05:33:38首次收录:2026/10/12 05:33:38原始来源
language
Cuda
stars
3967
forks
527
license
Apache-2.0
homepage
https://arxiv.org/abs/2410.02367
source_updated_at
2026-10-11T14:47:54Z

详细说明:

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

描述来源标记:原始/编辑整理

如何使用:

在 GitHub 打开仓库 thu-ml/SageAttention,按 README 安装与使用。Stars:3967。

同类对比:

词条关系简介选择理由
thu-ml/SageAttention当前—基准对照
LangChain相关The agent engineering platform.与「thu-ml/SageAttention」相关,可按场景对照选用
LlamaIndex相关LlamaIndex is the document processing platform for AI与「thu-ml/SageAttention」相关,可按场景对照选用
MCP 官方示例服务器集合相关Model Context Protocol 官方维护的参考 MCP Server 集合。与「thu-ml/SageAttention」相关,可按场景对照选用
ComfyUI相关AI 相关工具/产品:ComfyUI。与「thu-ml/SageAttention」相关,可按场景对照选用
AutoGen相关AI 相关工具/产品:AutoGen。与「thu-ml/SageAttention」相关,可按场景对照选用
NousResearch/hermes-agent相关The agent that grows with you与「thu-ml/SageAttention」相关,可按场景对照选用
firecrawl/firecrawl相关Supercharge your AI agents with data from the web and beyond. Building the library for superintelligence. 🔥与「thu-ml/SageAttention」相关,可按场景对照选用
huggingface/transformers相关🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training. 与「thu-ml/SageAttention」相关,可按场景对照选用

继续探索:

基于知识图谱关系,发现相关开源项目与跨分类内容。