Bisect AOTI compile duration upward: 600 → 1000 1e7f8dd moose Claude Opus 4.7 (1M context) commited on 26 days ago
Lower AOTI compile duration 1200 → 600 (cap probe) b7783cd moose Claude Opus 4.7 (1M context) commited on 26 days ago
Lower AOTI compile duration 1500 → 1200 (cap probe) e8d12ec moose Claude Opus 4.7 (1M context) commited on 26 days ago
Fix device mismatch during AOT compilation warmup c88fe83 moose Claude Opus 4.5 commited on Dec 28, 2025
Revert "Fix OOM by deferring model loading and AOT compilation to runtime" 31c7abd moose commited on Dec 28, 2025
Fix OOM by deferring model loading and AOT compilation to runtime bc7c2dd moose Claude Opus 4.5 commited on Dec 28, 2025