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

STATUS.md: validé/limites pour integrateur

This commit is contained in:
Richard Loyer 2026-05-24 22:17:42 +02:00
parent 9675280760
commit f877902903
1 changed files with 3 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
dist/STATUS.md vendored Normal file
View File

@ -0,0 +1,3 @@
# Statut & limites (24/05) — à lire avant intégration
VALIDÉ device: load 4B/9B HTP, prefill NPU + decode CPU, generate() bout-en-bout (test_native), 9B>4B qualité, cascade 9B+Guard4B 7.5GB. RAM -30% vs pte.
LIMITES: (1) prefill GDN-HTP non bit-exact (default OFF, GGML_HEXAGON_GDN_PREFILL=1 opt-in); decode CPU = exact. (2) TTS Talker GGUF chargeable mais audio e2e PAS validé (=JNI app). (3) generate() = prompt brut, app doit appliquer template chat + reasoning off. (4) STT inchangé ORT-QAIRT. (5) 30B/35B OOM. NON figé prefill>55 (HVX qf32 3-5sem).