Chantier 2 clos: CONT low-ROI (opsync gonfle, reshape structurel) -> 3 prioritaire

This commit is contained in:
Richard Loyer 2026-05-24 17:17:24 +02:00
parent 104c6b4ad5
commit ae842df5ae
1 changed files with 2 additions and 0 deletions

2
docs/02_cont_overhead.md Normal file
View File

@ -0,0 +1,2 @@
# Chantier 2 — CONT/copies prefill: ROI faible, non engage
416 CONT = ~17/couche GDN (reshape/permute DeltaNet, qwen3next.cpp). 4.6s mesure OPSYNC=serialise; async (defaut) les copies recouvrent le compute -> part reelle << gonfle. cont l.165 = TODO upstream (CUDA non-contig unary), pas removable safe. prefill deja 46-50, pas de bottleneck CONT prouve. Decision: low-ROI, skip. Le vrai gain prefill = GDN sur HTP (chantier 3), pas tailler des copies.