c111b3a9b0
Phase 01 (LLM Tuning): - Gemma4 26B: 74.65 t/s (fast) - Qwen 35B: 61.62 t/s (balanced) - Gemma4 31B: 16.0 t/s (deep-coder) - Qwen 27B: 16.7 t/s (deep-logic) - Qwen 122B: 8.95 t/s (ultra, GPU 1 only) Phase 02 (API Engine): - FastAPI reverse proxy on port 8000 - /engine/switch hot-swap with 503 protection - config/engine_models.json as single source of truth - Replaced 4 individual .bat files with unified engine File cleanup: - scripts/ 85 files -> 9 + _archive/ - Root .bat files -> _archive/
30 lines
751 B
Batchfile
30 lines
751 B
Batchfile
@echo off
|
|
chcp 65001 >nul
|
|
echo =========================================================
|
|
echo Qwen3.5 122B-A10B API Server (Final Optimized)
|
|
echo [CORE] --cpu-moe + -ngl 999: WDDM swap 완전 회피
|
|
echo [TUNED] -t 4: CPU 캐시 경합 최소화 (10.83 t/s)
|
|
echo [TUNED] mmap on: 메모리 매핑 활성 (no-mmap보다 빠름)
|
|
echo [VRAM] 5.4GB / 12GB (54%% 여유)
|
|
echo =========================================================
|
|
echo.
|
|
|
|
llama_bin_run\llama-server.exe ^
|
|
--model models\Q4_K_M\Qwen3.5-122B-A10B-Q4_K_M-00001-of-00003.gguf ^
|
|
-ngl 999 ^
|
|
--cpu-moe ^
|
|
-c 2048 ^
|
|
-np 1 ^
|
|
-fa on ^
|
|
--cache-type-k q4_0 ^
|
|
--cache-type-v q4_0 ^
|
|
-ub 256 ^
|
|
-b 1024 ^
|
|
-t 4 ^
|
|
--mlock ^
|
|
--prio 2 ^
|
|
--port 8000 ^
|
|
--host 0.0.0.0
|
|
|
|
pause
|