Commit Graph

25 Commits

Author SHA1 Message Date
e47bbf3e76 Compare Frontier rank under explicit objectives 2026-07-15 23:32:07 +08:00
ef5c17e6ec Use writable fleet root on dash0 2026-07-15 23:20:01 +08:00
91869c531c Add blind Qwen235B fleet job 2026-07-15 23:17:43 +08:00
33c91109d7 Require batched Frontier EP equivalence 2026-07-15 22:45:40 +08:00
1fabea26ef Batch exact Frontier EP lane predictions 2026-07-15 22:44:10 +08:00
dadcdbd351 Warm community rank runs before measurement 2026-07-15 22:32:40 +08:00
5dbcd9f51c Parallelize independent Frontier configs 2026-07-15 22:11:44 +08:00
ae933f2256 Record Frontier cache equivalence evidence 2026-07-15 22:10:36 +08:00
5fadc86edd Cache exact Frontier EP lane predictions 2026-07-15 22:07:40 +08:00
ed500fa016 Add fixed-cohort Frontier rank protocol 2026-07-15 21:47:25 +08:00
ca88c06d06 Preserve trace arrivals in Frontier grid 2026-07-15 21:08:19 +08:00
6307620e20 Prevent client-side clipping in vLLM grid 2026-07-15 20:36:10 +08:00
2f730417e0 Revert "Preserve global token count in Frontier EP profiles"
This reverts commit d64f12eb5b.
2026-07-15 20:26:27 +08:00
d64f12eb5b Preserve global token count in Frontier EP profiles 2026-07-15 20:23:21 +08:00
76e0de33ee Model critical EP lane for Frontier prefill 2026-07-15 20:15:09 +08:00
9275619984 Add reproducible community vLLM prefill grid 2026-07-15 20:00:41 +08:00
584af7b253 Fingerprint Frontier source snapshots 2026-07-15 19:41:07 +08:00
a6f101b09e Add reproducible Frontier prefill grid freeze 2026-07-15 19:39:30 +08:00
a54f69352d Add Frontier MoE gating context closure 2026-07-15 19:28:39 +08:00
c71f379110 Add Qwen235B NCCL profile runner 2026-07-15 19:10:52 +08:00
97e66ae276 Add validated Frontier profile assembly 2026-07-15 19:04:40 +08:00
e6e6fef41a Skip dense FFN surrogate for MoE profiles 2026-07-15 18:51:52 +08:00
c9a5f0a0c3 Profile exact vLLM MoE serving entrypoint 2026-07-15 18:41:23 +08:00
d9df3003dd Match in-place vLLM MoE serving path 2026-07-15 18:37:26 +08:00
684a2de413 Align Frontier FP8 profiling with vLLM runtime 2026-07-15 18:34:13 +08:00