Intel ARC PRO B70 本地部署大模型Qwen3.6 35B测试，速度能到40token/s, 测试平台:
CPU: i7 10700
内存：32G DDR4
显卡:Intel ARC PRO B70 32G
启动脚本:
@echo off
set LLAMA=E:\AI\llama_cpp\llama-server.exe
set MODEL=E:\AI\llama_cpp\Qwen3.6-35B-A3B-GGUF\Qwen3.6-35B-A3B-UD-Q6_K_XL.gguf
set MMPROJ=E:\AI\llama_cpp\Qwen3.6-35B-A3B-GGUF\mmproj-F32.gguf
set PORT=8080
set CTX=16384
set GPU=56
set BATCH=288
set UBATCH=288
set FLASH_ATTN=auto
set HOST=0.0.0.0
echo Starting llama-server with optimal parameters...
"%LLAMA%" ^
--model "%MODEL%" ^
--mmproj "%MMPROJ%" ^
--ctx-size %CTX% ^
--n-gpu-layers %GPU% ^
--batch-size %BATCH% ^
--ubatch-size %UBATCH% ^
--host %HOST% ^
--port %PORT% ^
--flash-attn %FLASH_ATTN%
echo Server started. Press Ctrl+C to stop.
pause
