Kazeia-engine/dist/lib
Richard Loyer c166391f49 dist: routage par structure (is_hybrid) — 3.5->option C, dense->CPU robuste
Détection au load via llama_model_is_hybrid/is_recurrent (structure, pas string d'archi):
- hybride GDN (qwen3.5/qwen3next) -> option C (prefill HTP/decode CPU), préservé+validé.
- dense (qwen3, etc.) / pas de HTP -> CPU pur (universel, robuste).
Charge l'instance CPU d'abord (détection + universel), n'ajoute l'instance HTP que pour l'hybride.

Diagnostic dense-HTP: le prefill dense sur HTP CRASHE (dspqueue_read 0x2e) sur prompt réaliste
(reproduit aussi en llama-cli -dev HTP0 -fa 1; le "ça marche" antérieur = prompt minuscule sans fa,
cas chanceux). Bug backend HTP attention dense multi-token, non corrigé. Dense a un .pte (451) de
toute façon -> dense=CPU robuste. Validé test_jni_native: dense (CPU) + 3.5 (option C) cohérents.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-27 12:30:43 +02:00
..
libc++_shared.so 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 +02:00
libggml-base.so 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 +02:00
libggml-cpu.so 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 +02:00
libggml-hexagon.so dist: prefill HTP corrigé (SSM_CONV gate backend), JNI option C sans OPFILTER 2026-05-27 10:39:08 +02:00
libggml-htp-v68.so 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 +02:00
libggml-htp-v69.so 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 +02:00
libggml-htp-v73.so 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 +02:00
libggml-htp-v75.so 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 +02:00
libggml-htp-v79.so 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 +02:00
libggml-htp-v81.so 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 +02:00
libggml.so 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 +02:00
libkazeia_engine.so dist: routage par structure (is_hybrid) — 3.5->option C, dense->CPU robuste 2026-05-27 12:30:43 +02:00
libllama-common.so 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 +02:00
libllama.so 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 +02:00