This website requires JavaScript.
5cd811b550
binaires .pte : libkazeia_pte.so + runtime QNN 2.42 (dist/lib-pte) + source JNI
2026-06-18 22:58:50 +0200
f68ecb4573
chantier LLM dispatch : .pte NPU + GGUF CPU, factory LlmLoader + doc dev + diag bug version QNN
2026-06-18 22:51:16 +0200
3775bbc37c
chantier TTS CosyVoice3 + verdict R&D NPU
2026-06-16 21:48:30 +0200
655d65a655
chantier RAG #2 : modèle FR tranché (e5-small voie a) + .so CPU-only + patch converter
2026-06-09 22:46:02 +0200
5cc8c22fc9
chantier RAG #1 : entrypoint embeddings (texte -> vecteur poolé) pour le RAG mobile
2026-06-09 22:02:30 +0200
f4c6e1290c
chantier B engine #3 : TTS lib audible + E2E in-app validé
2026-06-03 22:37:47 +0200
e55cd175c8
chantier B engine #2 : bloqueur SELinux découvert + guide rebuild CPU-only pour dev
2026-06-02 23:12:38 +0200
8c1e83be18
chantier B unification #1 : engine 3/3 prêt pour intégration (STT + TTS + LLM)
2026-06-01 15:46:20 +0200
4ad3b3716d
chantier B STT #6 : 2 bugs critiques fixés grâce à l'A/B in-app du dev
2026-06-01 12:19:33 +0200
dbd4f3d944
chantier B STT #5 : Action 1 dev — patch QNN options pour la prod gelée
2026-06-01 09:42:27 +0200
c9e198332d
chantier B STT #4 : audit dev pris en compte — options QNN explicites + bench apples-to-apples
2026-06-01 09:08:22 +0200
970f731e01
chantier B STT #3 : decoder zero-copy + sanity vocab + checklist dev
2026-05-31 22:26:40 +0200
8f6253204c
chantier B STT #2 : S2 + S3 livrés (stt_engine ORT QNN + JNI + façade Kotlin + doc)
2026-05-31 22:15:58 +0200
4b43d9e3d7
chantier B STT #1 : S1 unification engine (mel partagé + skeleton stt + VAD C++)
2026-05-31 19:19:28 +0200
82d206137f
chantier B TTS #8 : speaker encoder embedded + clonage in-process
2026-05-31 15:54:29 +0200
e42e8f8464
SE.1+SE.2+SE.3 partiel : speaker encoder pipeline POC bout-en-bout
2026-05-31 11:37:14 +0200
19ac123afa
TU sampler tuning : top_p 1.0->0.95 + rep_penalty 1.05->1.10 nouveaux défauts
2026-05-31 11:18:25 +0200
c332943181
Decoder chraac en sous-process : -20% decoder, -9% pipeline, talker intact
2026-05-30 18:14:09 +0200
2809b54943
CH chraac-llama : -18% RTF (2.43 -> 2.04) via swap ggml source
2026-05-30 12:38:47 +0200
d6acf378d5
POC streaming TTFB ÷4-6 (KZTTS_STREAM_CHUNK=K)
2026-05-29 22:56:14 +0200
27d893c73f
Vulkan Adreno 830 testé : neutre au mieux, régresse au pire (commit aussi le code opt-in)
2026-05-29 22:28:16 +0200
a5d591a07b
doc: PERF_CPU_OPTIMS.md — bilan optims CPU, RTF 2.93->2.45
2026-05-29 17:30:54 +0200
9a0c63f164
T2.1 CP heads NEON 16-way: CP -33%, pipeline -16%, bit-exact
2026-05-29 17:26:54 +0200
101d1cd7ed
tests infra HTP/HMX : verdict, le HTP n'est pas le bon outil pour TTS
2026-05-29 16:40:56 +0200
48c66c03e1
chantier F (CP HMX) : path sched livré, gain absent (overhead per-call)
2026-05-29 16:03:41 +0200
bdd528ed25
chantier BigVGAN HMX Phase 2b: refactor sched-friendly + limite backend découverte
2026-05-29 14:05:35 +0200
9de6b76838
chantier BigVGAN HMX Phase 0+2a: audit + chaîne load HTP validée
2026-05-29 11:51:08 +0200
1c25c975dc
doc: PERF_ANALYSIS.md — diagnostic post-session, leviers triés ROI
2026-05-28 22:43:23 +0200
930f3c8c1f
chantier B TTS #11 : libkazeia_tts.so + Kotlin wrapper, JNI validé bout-en-bout
2026-05-28 21:54:21 +0200
b27c8ef403
chantier B TTS #10 : tokenizer Qwen3 BPE via llama_tokenize, texte arbitraire
2026-05-28 21:21:53 +0200
1203d7ce5d
chantier B TTS #9 : CP weights f16 conservés, CP -52% cumulé, bit-exact
2026-05-28 21:14:21 +0200
7626131d50
chantier B TTS #8 : KV cache CP -40% par frame, bit-exact
2026-05-28 19:51:22 +0200
d20f2b297b
chantier B TTS #7 : sampler HF-style top_k+top_p+rep_penalty (générique)
2026-05-28 16:55:05 +0200
9ef5e11b53
chantier B TTS #6 : P3.4+P3.5 = pipeline complet 100% tablette bout-en-bout
2026-05-28 15:25:00 +0200
746bbe77cd
chantier B TTS #5 : P3.3 = vrai CP runner intégré (bit-exact)
2026-05-28 15:11:07 +0200
7a998dec6b
chantier B TTS #4 : P3.1+P3.2 = x_vector + prefill_embeds depuis input_ids (bit-exact)
2026-05-28 15:05:23 +0200
2e0b145b51
chantier B TTS #3 : orchestration ggml-side bit-exact + WAV E2E sur tablette
2026-05-28 14:25:18 +0200
056dc2831a
ql: bump pour qwen3 dense IMROPE (ql:6e0edba)
2026-05-28 13:54:27 +0200
12df0d0ba3
chantier B TTS #2 : fix M-RoPE positions en embeds-mode + validation bit-exact
2026-05-28 13:54:03 +0200
50814f3bca
chantier B TTS #1 : API JNI embeds-only + M-RoPE qwen3 dense
2026-05-28 13:37:17 +0200
9af47bf388
chantier dense-HMX #4 : 0x2e décodé = AEE_ECONNRESET (reset watchdog CDSP), pas une corruption de valeurs
2026-05-27 19:20:28 +0200
97180b29a6
dist: audit intégrabilité HANDOFF — corrige passages périmés
2026-05-27 18:55:16 +0200
b350c2be73
chantier dense-HMX #3 : uninit VTCM matmul écarté (zéro-init ne corrige pas). Source -inf non-det = amont, instrumentation DSP requise.
2026-05-27 16:54:58 +0200
464d4e69c0
chantier dense-HMX: avancement #2 (replay) — root = fault HMX non-déterministe, instrumentation DSP requise (bloquée sans root). Tools committés.
2026-05-27 14:35:14 +0200
224b375a10
chantier dense-HMX: avancement — ni magnitude, ni NaN, ni shape pure (value/context-spécifique)
2026-05-27 14:09:52 +0200
5aa4457567
chantier: instrumentation crash dense-HTP-HMX + .pte retiré (engine seul runtime)
2026-05-27 13:29:32 +0200
0407ac8b71
dist: ROOT CAUSE crash dense-HTP = matmul HMX (fp16) ; fix = HVX pour le dense
2026-05-27 12:42:15 +0200
c166391f49
dist: routage par structure (is_hybrid) — 3.5->option C, dense->CPU robuste
2026-05-27 12:30:43 +0200
eeffbc350f
dist: JNI générique (auto-détecte archi) — tout LLM, sans dégrader la 3.5
2026-05-27 12:11:44 +0200
771d64a998
dist: multi-tour propre (generateRaw + ChatSession) + JNI validé end-to-end natif
2026-05-27 11:18:20 +0200
6935089459
dist: libkazeia_engine.so rebuild option C (paquet cohérent), notes rebuild MAJ
2026-05-27 10:45:47 +0200
2699b631a3
dist: prefill HTP corrigé (SSM_CONV gate backend), JNI option C sans OPFILTER
2026-05-27 10:39:08 +0200
2c89ec01a2
dist: JNI option C câblé (prefill HTP / decode CPU), validé multi-tour
2026-05-27 10:24:55 +0200
501abb67e5
dist: prefill HTP RÉCUPÉRABLE — charabia = 1 op (SSM_CONV) cassée, fix OPFILTER
2026-05-27 10:12:37 +0200
2256c077c9
dist: prefill HTP = CASSÉ (sortie charabia), prefill obligatoirement CPU
2026-05-27 09:42:59 +0200
c9d4e625eb
dist: métriques vérifiées batterie pleine — KV f16 (pas q8_0), q35-lmq4 HTP=189
2026-05-27 09:23:57 +0200
fac50d69ce
dist: bandeau HANDOFF source de vérité sur README/INTEGRATION (9B historique)
2026-05-26 23:34:15 +0200
9417aba6cb
dist: handoff intégration aligné décisions 26/05 (modèle q35-lmq4, perf, prompt)
2026-05-26 23:33:41 +0200
21784d52de
eval: verdict (assisté, à valider) -> ship Qwen3.5-lmq4
2026-05-26 23:23:00 +0200
777c6979c1
eval: kit qualité A/B aveugle Qwen3.5-lmq4 vs dense (decode CPU, template, reasoning off)
2026-05-26 22:24:23 +0200
a7c0e96793
RAPPORT §7: piste kernel GDN fermee — GDN-compute != goulot prefill (fork ql)
2026-05-26 21:46:05 +0200
dd29169672
GDN chunkwise P1.2: VTCM state prefill=neutre, prouve compute-bound 3S2/tok -> gain=WY matmul (palier dur). oracle 28/28, pp98
2026-05-26 16:30:08 +0200
f638bdbd94
GDN chunkwise P1.1: chunk C=64 refactor neutre, oracle 28/28+pp98 OK, squelette pret pour intra-chunk batch
2026-05-26 16:15:55 +0200
837d9de766
GDN chunkwise: doc P1 math+structure+gate (refactor chunks C=64 phase1)
2026-05-26 16:13:34 +0200
aa363e652f
GDN chunkwise P0: oracle test-backend-ops 28/28 HTP0 (gate régression), baseline pp512=93, boucle build live; P1 chunk-f32 en cours
2026-05-26 16:04:58 +0200
de9543de2a
Chantier4 GDN chunkwise: plan grounded metrics, kernel pp_thread sériel localisé, HMX core_dot_chunk_fp16 réutilisable, build htp-v79 incrémental OK
2026-05-26 15:55:25 +0200
b2fc564f04
Engine sur fork qualcomm: decode dense 15.5>pte, 3.5 ~11, KVq8+t4+fa+lm_head-Q4; prefill 93/285; RAG cross 24-28k
2026-05-26 15:53:14 +0200
1252f32bd3
Rapport R&D complet: synthese session, perf, cascade, #277 , briques, reste a faire
2026-05-25 22:27:34 +0200
9b3262a8aa
R&D: battre pte=lm_head HMX INT4+spec decode+ctx binary; mur=lm_head CPU
2026-05-25 18:46:28 +0200
025bdd87bd
Briques: 2B 13, 4B 7, 9B 2.1; 2B sweet spot
2026-05-25 18:34:16 +0200
cbc379f9e5
Chiffres stables r3: 4B 9.8 9B 5.2 (r1 etaient bruit); engine tient, pte 4B devant
2026-05-25 17:41:08 +0200
8cc9286f70
Correction: 9B CPU ~4.4 (t8 contention), <8B pte; 4B seul GGUF viable
2026-05-25 15:54:46 +0200
e462d7ce27
9B cible: 1-coeur in-app=0.16, t8=8.4; pte8B repli; merge=affinity 8 coeurs
2026-05-25 13:04:21 +0200
7aae1f5fc7
Final: pte 4B 15/2.2GB > GGUF 0.32; merge=t8 mais pte reste mieux 4B; gguf utile=gros modeles
2026-05-25 12:21:04 +0200
cdb34a65f1
0.32 = 1 thread + HTP residuel; fix t8+ngl0 = 14tok/s
2026-05-25 11:18:38 +0200
23f2c9fcfc
#277 : dist=static ngl0 OK; crash dev=stale shared so + ancienne signature 1-arg
2026-05-25 10:44:53 +0200
9b62420089
v4 fix #277 : ngl0 CPU decode 14tok/s, 4s/tour, usable
2026-05-25 10:18:31 +0200
7a5780ca1d
Bridge v3 utilisable: thinking-off ChatML, sys+usr, cap; static intact
2026-05-25 09:41:04 +0200
fe6826a3be
Integration #277 : hang=thinking non bride, fix budget0+cap; 4B>9B CPU
2026-05-25 09:38:25 +0200
76f21a21cb
Bridge STATIC autonome (42MB): zero NEEDED partage, fix collision TTS libllama, valide device
2026-05-24 22:43:45 +0200
f877902903
STATUS.md: validé/limites pour integrateur
2026-05-24 22:17:42 +0200
9675280760
GDN chunkwise: HMX fp16 bloque bit-exact, HVX qf32 3-5sem; reste 55/15 CPU exact
2026-05-24 22:12:47 +0200
1a6637bd67
TTS: Talker GGUF chargeable, audio e2e=JNI app non valide shell; GDN agent en cours
2026-05-24 22:12:23 +0200
c5c2738281
Valide device: generate() bout-en-bout OK (test_native), bridge pret integration
2026-05-24 22:05:44 +0200
7dbf38db97
Doc complete: README+MODELS+PERF+PITFALLS, build nettoye
2026-05-24 21:59:56 +0200
15b9baf6b2
JNI bridge complet: generate() prefill+decode, build arm64 OK (libkazeia_engine.so 209K)
2026-05-24 21:58:00 +0200
0f92fa723f
Dist Kazeia-Engine: libs arm64+headers+JNI bridge+INTEGRATION/TTS doc; STT reste ORT
2026-05-24 21:55:01 +0200
1024167bdc
Empreinte: engine GGUF 2.4/5.0 vs pte 3.3/5.9, -15-30%
2026-05-24 21:18:16 +0200
9f7bc1e6c3
Comparatif FR: 9B>4B (psy juste vs derive role), engine OK budget0
2026-05-24 21:12:10 +0200
431865f07f
Fix cascade: reasoning-budget 0 (engine OK, FR coherent); A/B = JNI
2026-05-24 20:15:41 +0200
25b1344c19
Cmp 4B/9B: 9B+riche, mais eval FR shell impossible, requiert JNI
2026-05-24 19:50:24 +0200
db6c063705
Cascade 9B+Guard4B/8B: speed+RAM mesures, reco Guard4B; eval texte = JNI
2026-05-24 19:30:56 +0200
7f9dda4a75
Cascade 9B+Guard4B tient (7.5GB), Guard8B ok aussi; verdict gros modeles
2026-05-24 19:21:08 +0200
f82fa61281
9B: 41/8.4 5GB tient; 30B/35B OOM. Plafond=9B+4B parallele, MoE hors-jeu
2026-05-24 19:13:20 +0200
47e75d74a4
Prefill cap = recurrence serielle, HVX inutile; gain=chunkwise GDN matriciel (weeks)
2026-05-24 18:55:49 +0200
e1a6d4e764
Prefill: HMX/batch satures, GDN scalaire ancre, plafond ~103 via HVX chunkwise
2026-05-24 18:29:46 +0200
ce5d91946e
Comparatif: Q3-4B 103/17, Q3.5-4B 55/15, ExecuTorch 14-21. Engine egale decode sans pte
2026-05-24 18:19:27 +0200
eb72df4767
Mesure split ref: prefill HTP 55 / decode CPU 15 Qwen3.5-4B, end-to-end ~13s
2026-05-24 18:08:39 +0200
1b64bbb7eb
Clarif: gen lente = decode-NPU cli, pas GDN; split prefill-NPU/decode-CPU = role JNI
2026-05-24 18:06:40 +0200
8c793a22b5
FIX GDN prefill HTP: GQA broadcast h%heads, charabia resolu, pp512 58 t/s (+16%)
2026-05-24 17:57:20 +0200
Powered by TurnKey Linux .
Richard Loyer