about things notes.zzstoatzz.io
notes

ai/models/jev: open reproductions, kev-4b on modal and its latency causes master

Leaderboard of open System One reproductions, the kev-4b modal spike, and the measured latency breakdown: close-read padding, per-row state recompute, gpu, modal ingress. Corrects the earlier claim that the causal_conv1d wheel loaded; it needs glibc 2.32. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01RbTFzpWhDzcb9g4AwRUvUE


+143 -1
2 changed files