Commit Graph

13 Commits

Author SHA1 Message Date
coolboy
dc8365c215 fix: prevent stale capacity poisoning and deepen recent-only refill scans
Keep ordinary capacity races from lowering persistent per-account limits.
Deepen recent-only discovery within the existing age and quality gates.
Add offline regressions for recovery and reserved-slot refill.
2026-10-02 21:24:08 +08:00
CoolBoy
6154dc738f fix: restore accuracy-first route admission 2026-09-22 15:26:42 +08:00
CoolBoy
4afce879dc feat: calibrate account capacity weekly 2026-09-20 13:03:35 +08:00
CoolBoy
75c034d588 fix: recover eligible model GPU routes 2026-09-20 10:47:22 +08:00
CoolBoy
ff73768537 feat: compact durable logs and refine framework routing 2026-09-04 10:30:21 +08:00
CoolBoy
91d1d3d87d feat: make model age admission-only 2026-08-12 10:18:09 +08:00
CoolBoy
2065ad6abc feat: derive recent queue reserve from account capacity 2026-08-11 01:43:40 +08:00
CoolBoy
f12d96b138 feat: reserve queue capacity for recent models 2026-08-11 01:29:54 +08:00
CoolBoy
38ab25fc3c refactor: remove LLM from online decisions 2026-08-11 00:59:40 +08:00
CoolBoy
8e68c6f611 fix: refill from expanded candidate history 2026-08-02 17:45:21 +08:00
CoolBoy
eab5ab6dce feat: add adaptive GPU scheduling 2026-08-02 16:59:44 +08:00
CoolBoy
0e42ef73dc fix: replace duplicate submissions while refilling queues 2026-08-02 15:44:28 +08:00
CoolBoy
0cabc43abf fix: coordinate concurrent account capacity filling 2026-08-01 16:50:50 +08:00