HNHacker News
TopNewBestAskShowJobs

luxiedge

2 karma · joined December 8, 2025

submissionscomments

Constant 14ms attention: 512→524K tokens (24.5x faster than FlashAttention)

github.com·1 pts·luxiedge·
1

Show HN: O(1) memory attention – 512K tokens in 3.85 GB (eval binary)

github.com·1 pts·luxiedge·
0

Why do math libraries produce different results across platforms?

github.com·3 pts·luxiedge·
3