Kazeia
/
Kazeia-engine
3
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Clarif: gen lente = decode-NPU cli, pas GDN; split prefill-NPU/decode-CPU = role JNI

This commit is contained in:
Richard Loyer 2026-05-24 18:06:40 +02:00
parent 8c793a22b5
commit 1b64bbb7eb
1 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
CLAUDE.md
View File

@ -63,3 +63,6 @@ Sweep ngl: 0=18.7/8=19/16=18/24=23/32=33/99=45 -> monotone, full offload gagne.
GATED_DELTA_NET ABSENT du profil HTP = GDN 100% CPU confirme. HTP top: CONT(copies) 4.6s, SSM_CONV 2.9, MUL_MAT 2.9, REPEAT, CPY. opsync gonfle, GDN_cpu = residu sale (~total-Reste), facteur 2.5 non leve net. Fait dur: 24/32 couches GDN hors NPU = sujet B-2. CONT/copies = overhead a tailler aussi. prefill 48.9. GATED_DELTA_NET ABSENT du profil HTP = GDN 100% CPU confirme. HTP top: CONT(copies) 4.6s, SSM_CONV 2.9, MUL_MAT 2.9, REPEAT, CPY. opsync gonfle, GDN_cpu = residu sale (~total-Reste), facteur 2.5 non leve net. Fait dur: 24/32 couches GDN hors NPU = sujet B-2. CONT/copies = overhead a tailler aussi. prefill 48.9.
## Chantier1 clos: prefill pp2048=46 sain, hang cli only (lib OK pour Kazeia/JNI) ## Chantier1 clos: prefill pp2048=46 sain, hang cli only (lib OK pour Kazeia/JNI)
## Note: gen lente cli = decode sur HTP (pas le fix)
-dev HTP0 garde decode sur NPU = 3-4 t/s + reasoning 200tok = 0.6 t/s percu. Vrai: prefill HTP 58 / decode CPU 13 (-dev none). cli mono-device, ne splitte pas. Cible = JNI prefill HTP -> bascule decode CPU. 0.6 t/s = artefact bench, pas perf.