Local Model Leaderboard Watch
Search the build
-
9661d342026-09-20 chooser 2026-09-20: pick Qwen3-30B-A3B-Instruct-2507-4bit (only real weights on disk) -
9a759642026-09-17 auto-data-snapshot: 2026-09-17T05:03:29 (1 data files) — data/chosen-model.txt -
da185cb2026-09-14 auto-data-snapshot: 2026-09-14T05:09:49 (1 data files) — data/chosen-model.txt -
373cb152026-09-13 auto-data-snapshot: 2026-09-13T05:22:51 (1 data files) — data/chosen-model.txt -
831b6cb2026-09-11 auto-data-snapshot: 2026-09-11T08:57:34 (1 data files) — data/chosen-model.txt -
c7d4b712026-09-10 Record independent verification of adopted model guard -
156aa462026-09-10 Validate tensor shape byte lengths before allowing model load -
6fe80962026-09-10 Refuse model loads when local weight shards are incomplete -
4c167962026-09-03 auto-data-snapshot: 2026-09-03T04:44:57 (1 data files) — data/chosen-model.txt -
9d789902026-09-02 auto-data-snapshot: 2026-09-02T05:17:25 (1 data files) — data/chosen-model.txt -
d1aa0212026-08-25 auto-data-snapshot: 2026-08-25T05:12:48 (1 data files) — data/chosen-model.txt -
cccfb0f2026-08-19 auto-data-snapshot: 2026-08-19T04:49:28 (1 data files) — data/chosen-model.txt -
bc626c52026-08-18 battle: 80B-A3B-4bit 12/12@21tok/s vs 35B 11/12@55tok/s — head-to-head results + harness fixes -
524d2a52026-08-18 auto-data-snapshot: 2026-08-18T09:42:51 (1 data files) — battle/80b-watch.log -
6963c732026-08-18 battle: eval harness + 35B baseline (11/12, 55.2 tok/s, longctx pass) -
a52578d2026-08-18 daily chooser 2026-08-18: chose Qwen3.6-35B-A3B-8bit; correct upgrade→Qwen3-Next-80B-A3B-4bit (44.8GB fits), drop gpt-oss-120b (too big) -
0747d9f2026-08-17 purge remaining 90GB refs -> 55GB weights ceiling everywhere -
dddd5832026-08-17 tighten fit ceiling to ~55GB weights (gpt-oss-120b 70.6GB failed placement 8/17) -
c96a2b62026-08-17 daily 4:45am + choose-best-downloaded-model-for-the-day (wrapper loads if on disk; downloads stay gated) -
7070cc02026-08-17 hourly local-model leaderboard watcher (no-Bash agent, recommend-only, gated)
Authors
- auto-commit-fleet9
- Steve7
- steve4
Agents used
Skills used
- /chosen-model8