Flash-MSA: Accelerating Million-Token Training with Sparse Attention Kernels

rawsh·Hacker News·Community·

Flash-MSA: Accelerating Million-Token Training With Sparse Attention Kernels

Read full article →

Related Articles

Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k
systima · Hacker News · 5h ago
Irish datacenters now guzzle 23% of the country's electricity
Bender · Hacker News · 3h ago
Since Chromium 148, Math.tanh is now fingerprintable to link underlying OS
joahnn_s · Hacker News · 2h ago
Old and new apps, via modern coding agents
subset · Hacker News · 12h ago
AI boosts research careers but narrow the span of ideas explored: study
zaikunzhang · Hacker News · 10h ago