Package profile
flash-attn
- Summary: Flash Attention: Fast and Memory-Efficient Exact Attention
- Author: Tri Dao
- License: BSD License
- Homepage: https://github.com/Dao-AILab/flash-attention
- Source: https://github.com/Dao-AILab/flash-attention
- Number of releases: 76
- First release: 0.2.0 on 2022-11-15
- Latest release: 2.8.3.post1 on 2026-06-11
- Latest release size: 8.1 MB (sdist)
1 known vulnerabilityLatest: GHSA-7g5w-pq96-8c5w — flash-attention contains an insecure deserialization vulnerability in its checkpoint loading mechanism View all →
Dependencies
Flash-attn has 2 dependencies, 0 of which optional.Dependent packages
| Package | Optional | Group |
|---|---|---|
| crfm-helm | true | audiolm |
| flagembedding | true | finetune |
| fschat | true | train |
| mteb | true | flash-attention |
| nanotron | true | fast-modeling |
Similar packages
- lazy-object-proxyA fast and thorough lazy object proxy.
- fasttext-langdetect80x faster and 95% accurate language identification with Fasttext
- blinkerFast, simple object-to-object and broadcast signaling
- pebbleThreading and multiprocessing eye-candy.
- fastaifastai simplifies training fast and accurate neural nets using modern best practices
- qudidaQUick and DIrty Domain Adaptation
- transformersTransformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.