HN
Paper
All
Show
Ask
Jobs
Top
Today
Last 7 days
Last months
This year
Statistics
All
Show
Ask
Jobs
Top stories
Today
Last 7 days
Last months
This year
Statistics
Stories by
pidtom
Skipping 90% of KV dequant work speeds up LLM decode by 22%
1 points
pidtom
2026-03-27T14:59:40Z
github.com