|
|
d5bb9745f8
|
Reuse host-local caches for long-context replay
|
2026-07-24 01:50:17 +08:00 |
|
|
|
da480a8761
|
Extend serving rotary cache for code context
|
2026-07-24 01:14:15 +08:00 |
|
|
|
766f09d3ed
|
Enable gated long context real trace replay
|
2026-07-24 00:59:40 +08:00 |
|
|
|
7aed90f9e6
|
Support long code traces in Frontier replay
|
2026-07-24 00:57:08 +08:00 |
|
|
|
89d5ebbbc0
|
Add code trace max model length smoke
|
2026-07-24 00:49:12 +08:00 |
|
|
|
8462afa56f
|
Encode zero cacheable blocks explicitly
|
2026-07-24 00:40:15 +08:00 |
|
|
|
5a4011ca12
|
Exclude partial prompt blocks from prefix cache replay
|
2026-07-24 00:35:48 +08:00 |
|
|
|
ab952b47e7
|
Make synthetic trace block identities parent-sensitive
|
2026-07-24 00:11:48 +08:00 |
|
|
|
029c8991b6
|
Generalize trace remapping for code workloads
|
2026-07-24 00:09:50 +08:00 |
|
