Ettore Di Giacinto ae76d42a96 docs(paged): profile MTP graph reuse loss ...

Record Phase 16 nsys evidence that current MTP serving loses paged decode graph reuse and increases GPU work, explaining the Phase 15 serving regression.

Assisted-by: Codex:gpt-5

2026-07-01 02:32:49 +00:00

..

plans

docs(paged): profile MTP graph reuse loss

2026-07-01 02:32:49 +00:00

specs

docs(paged): gate MTP rollback safety

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