Prefrontal-AI/latent-sparse-attention

Code for Latent Sparse Attention: Towards Subquadratic Ultra-Long Context Attention

Python

362

5 contributors

Apache License 2.0

Fastest Enterprise AI Gateway for Routing, Governing, and Securing all your AI Traffic

Momentum

How fast Prefrontal-AI/latent-sparse-attention is gaining stars, and where it lands on Trendshift and GitHub Trending

Stars, forks and merged pull requests each day or month, lined up with the repository's rank on Trendshift Trending

Nov 1 – Nov 4, 2026

Stars +363 in 4 days

240

Forks +15 in 4 days

Merged PRs 0 in 4 days

Trendshift Trendingwhere the repository ranked each day it made the top 25

Not in the Trendshift top 25 in this period

Nov 1Nov 2Nov 3Nov 4

Stars, forks, merged pull requests and Trendshift Trending rank, by day
Day	Stars gained	Forks gained
Sun, Nov 1, 2026	+0	+0
Mon, Nov 2, 2026	+28	+4
Tue, Nov 3, 2026	+240	+11
Wed, Nov 4, 2026	+95	+0
Total	+363	+15

Recent discussions about this repository across the web

Nov 41 post
- The 22.7x in LSA's headline is one attention layer at 10M tokens, and the whole model gets less End to end on GLM-5.3-Flash at 10M, first token is 2.2x faster with context parallelism and 5.8x with…
  Read full on X, posted by @Adham__Khaled__
  @Adham__Khaled__ · x.com
Nov 21 post
- When training a Hybrid model from scratch, LSA tracks Full Attention's validation loss, but takes 1.24x less wall clock time (each attention layer trains 1.6× faster). No other sparse attention…
  Read full on X, posted by @jcrwhittington
  @jcrwhittington · x.com