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

Commit Graph

  • 31afd64eec GDN dump: FARF hors logcat (sink dedie req), lib nettoyee, B-2 tracke. Produit 50/13 pret Richard Loyer 2026-05-24 17:42:34 +0200
  • 6e63110f8a GDN prefill: bug=handoff multi-token (pas GQA), fix=dump bisection; defaut CPU sur Richard Loyer 2026-05-24 17:38:35 +0200
  • 33e848be3c Chantier 3: gate GDN-prefill leve = charabia, +11% INVALIDE, defaut CPU. Honnete: kernel a fixer Richard Loyer 2026-05-24 17:22:52 +0200
  • b82420fd46 Chantier 3 etape 1: GDN prefill HTP, gate leve, pp512 +11% (50->55) Richard Loyer 2026-05-24 17:20:09 +0200
  • ae842df5ae Chantier 2 clos: CONT low-ROI (opsync gonfle, reshape structurel) -> 3 prioritaire Richard Loyer 2026-05-24 17:17:24 +0200
  • 104c6b4ad5 CLAUDE: chantier1 clos Richard Loyer 2026-05-24 17:16:44 +0200
  • 024f3b8ed6 Chantier 1 clos: prefill pp2048 sain (46t/s), hang=cli only hors JNI Richard Loyer 2026-05-24 17:16:44 +0200
  • 17a78f1c6a Profil: GDN absent HTP (100% CPU confirme), CONT+SSM_CONV+MULMAT dominent; lib restauree Richard Loyer 2026-05-24 16:53:08 +0200
  • 3a5f735ab9 Optim prefill: ngl99/t8/b512=49.9 t/s (+11%), full offload, decode CPU Richard Loyer 2026-05-24 16:35:52 +0200
  • 627f9a80e2 Correctif: OPFILTER inexistant, isolation GDN_cpu = bench DEBUG; bugs+qualite requalifies Richard Loyer 2026-05-24 15:58:18 +0200
  • d05483b13a Bug cli = overflow buffer HTP gros prefill (2^44), pas init; qualite 4B OK Richard Loyer 2026-05-24 15:52:33 +0200
  • 753d8be2b0 pp1024 plat (saturation batchable), bornes GDN: cout B-2 non mesure, profiler bloque par cli Richard Loyer 2026-05-24 15:44:45 +0200
  • 4dd863ecb5 Verdict: GDN HTP decode-only (2284), prefill GDN fallback CPU = ancre. B chiffre. Richard Loyer 2026-05-24 14:49:39 +0200
  • eceacaba9a Mesures 4B: HTP prefill 44 / CPU decode 13.4, qwen35_compile HTP fonctionne Richard Loyer 2026-05-24 14:33:17 +0200
  • 2779babcab Rebuild upstream+hexagon: configure OK (NDK r27d, SDK 6.5, v79), build lance Richard Loyer 2026-05-24 13:01:16 +0200
  • 6aa4d06c95 Ambition haute: cible = Qwen3.5-4B near-ExecuTorch, prefill GDN sur HTP, decode CPU Richard Loyer 2026-05-24 12:57:37 +0200
  • 12217d3bbc Decision A: integration produit, base upstream master, pas de kernel GDN custom Richard Loyer 2026-05-24 12:54:28 +0200
  • 376b4a4a70 Angle revise: base upstream master (pas chraac), trou unique = prefill GDN Hexagon Richard Loyer 2026-05-24 12:52:08 +0200
  • cc5b033762 Engine cadrage: CLAUDE.md analyse R&D, verdict NPU-prefill/CPU-decode Richard Loyer 2026-05-24 12:34:14 +0200