wip: [01-llm-tuning] paused at task 1/3

start_gemma4_26b_api.bat

@@ -1,8 +1,30 @@
 @echo off
+chcp 65001 >nul
 echo =========================================================
-echo  Gemma4 26B-A4B API Server (Tuned for Max Speed)
-echo  [INFO] Tuning VRAM limit correctly to avoid WDDM swap (-ngl 22)
+echo  Gemma4 26B-A4B API Server (256K Context - Final Optimal)
+echo  [CORE] --n-cpu-moe 10: VRAM 12GB 최적화용 Expert 오프로드
+echo  [TUNED] -t 4 -ub 512: CPU 병목 방지 및 SWA 캐시 최적화
+echo  [PERF] Speed: ~30.9 t/s (1x RTX 3060)
 echo =========================================================
 echo.
-llama_bin_run\llama-server.exe --model models\gemma-4-26B-A4B-it-Q4_K_M.gguf -ngl 22 -c 4096 -np 1 -fa on --cache-type-k q4_0 --cache-type-v q4_0 -ub 512 -b 2048 -t 8 --mlock --prio 2 --port 8000 --host 0.0.0.0
+
+llama_bin_run\llama-server.exe ^
+  --model models\gemma-4-26B-A4B-it-Q4_K_M.gguf ^
+  -ngl 999 ^
+  --n-cpu-moe 10 ^
+  -c 262144 ^
+  -np 1 ^
+  -fa on ^
+  --cache-type-k q4_0 ^
+  --cache-type-v q4_0 ^
+  -ub 512 ^
+  -b 2048 ^
+  -t 4 ^
+  -tb 4 ^
+  --mlock ^
+  --prio 3 ^
+  --poll 50 ^
+  --port 8000 ^
+  --host 0.0.0.0
+
 pause