Ettore Di Giacinto 4d171e62bb docs(paged): reject MTP serving lever ...

Add the repeatable MTP serving A/B runner and record Phase 15 results showing current llama-server MTP regresses GB10 serving throughput despite passing inference gates.

Assisted-by: Codex:gpt-5

2026-07-01 02:29:28 +00:00

..

2026-06-30-gb10-parity-reopen.md

docs(paged): record W4A16 phase1 kill gate

2026-06-30 20:40:40 +00:00

2026-06-30-serving-nsys-phase6.md

docs(paged): scope phase7 serving candidates

2026-06-30 23:12:09 +00:00

2026-06-30-serving-source-phase7.md

docs(paged): reject weighted combine fusion candidate

2026-07-01 00:20:53 +00:00

2026-06-30-w4a16-kernel-shape-phase2.md

patches(paged): tune W4A16 grouped tile shape

2026-06-30 21:57:42 +00:00

2026-06-30-w4a16-metadata-phase1.md

patches(paged): mirror W4A16 packed metadata

2026-06-30 21:21:53 +00:00

2026-06-30-w4a16-scale-broadcast-phase3.md

docs(paged): record W4A16 scale broadcast rejection

2026-06-30 22:06:17 +00:00

2026-06-30-w4a16-shmem-pad-phase4.md

docs(paged): close W4A16 pad checklist

2026-07-01 01:58:22 +00:00

2026-06-30-w4a16-wq-pad-phase5.md

docs(paged): record W4A16 Wq padding rejection

2026-06-30 22:23:14 +00:00

2026-07-01-gdn-c32-slab-phase10.md

docs(paged): reject GDN C32 slab phase

2026-07-01 01:15:00 +00:00

2026-07-01-gdn-global-ai-prototype-phase13.md

docs(paged): reject GDN global Ai32 prototype

2026-07-01 01:51:53 +00:00

2026-07-01-gdn-m5-state-boundary-phase11.md

docs(paged): reject GDN M5 QS-early phase

2026-07-01 01:29:44 +00:00

2026-07-01-gdn-shared-ai-cost-model-phase12.md

docs(paged): reject GDN global Ai32 prototype

2026-07-01 01:51:53 +00:00

2026-07-01-mtp-draft-smoke-phase9.md

fix(paged): gate MTP backend sampling

2026-07-01 00:54:25 +00:00

2026-07-01-mtp-rollback-serving-gates-phase14.md

docs(paged): gate MTP rollback safety

2026-07-01 02:15:11 +00:00

2026-07-01-mtp-serving-throughput-phase15.md

docs(paged): reject MTP serving lever

2026-07-01 02:29:28 +00:00

2026-07-01-serving-ragged-moe-phase8.md

docs(paged): close ragged MoE dispatch shortcut

2026-07-01 01:57:45 +00:00