Commit Graph

2 Commits

Author SHA1 Message Date
Richard Loyer 5cd811b550 binaires .pte : libkazeia_pte.so + runtime QNN 2.42 (dist/lib-pte) + source JNI
- dist/lib-pte/ : libkazeia_pte.so (wrapper) + 7 libs QAIRT 2.42 (libqnn_executorch_backend
  + libQnnHtp/System/HtpPrepare/HtpV79Stub/HtpV79Skel/HtpNetRunExtensions) + qnn_llama_runner (oracle test).
- dist/jni/kazeia_pte_jni.cpp : source du wrapper JNI.
- docs : chemins recalés sur dist/lib-pte (au lieu de pte_prep_work hors repo).
- .pte (multi-Go) restent hors git (box.kazeia.com).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-18 22:58:50 +02:00
Richard Loyer f68ecb4573 chantier LLM dispatch : .pte NPU + GGUF CPU, factory LlmLoader + doc dev + diag bug version QNN
- LlmLoader.kt : chargeur agnostique (magic-byte GGUF/ET12), interface LlmEngine,
  GgufLlmEngine (CPU-i8mm, hybride Qwen3.5-4B) + PteLlmEngine (NPU, denses).
- LLM_INTEGRATION.md : doc dev complète (modeles, perf mesuree, integration, SELinux no-root,
  export .pte) ; remplace la v02/06 perimee (GGUF+ggml-hexagon).
- .pte denses prepares+valides on-device : Qwen3-4B 15.7 / Qwen3-8B 10.5 / Guard-4B 17.2 /
  Qwen2.5-7B 8.0 tok/s ; wrapper JNI libkazeia_pte parite bit-exacte.
- BUG_pte_inapp_qnn_version_mismatch.md : diag in-app err 5010 = backend QNN != 2.42 (pas le .pte) ;
  ancien rapport overflow int32 retracte (refute par test CLI 8B).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
2026-06-18 22:51:16 +02:00