NickCanCode

beellama launch script

Jun 5th, 2026
74
0
Never
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
Bash 0.70 KB | Software | 0 0
  1. S:\beellama.cpp-win\build-msvc-cuda-server\bin\llama-server ^
  2. --model A:\models\Qwen3.6-27B-Q2_K_MIXED-AutoRound-MTP.gguf ^
  3. --ctx-size 160000 ^
  4. --spec-draft-model "S:\LLMs\Qwen3.6-27B-DFlash-IQ4_XS.gguf" ^
  5.   --spec-type dflash ^
  6.   --spec-dflash-cross-ctx 1024 ^
  7. --spec-draft-ngl all ^
  8. --batch-size 2048 ^
  9. --gpu-layers all ^
  10. --cache-type-k q8_0 --cache-type-v q8_0 ^
  11. --spec-draft-type-k q8_0 --spec-draft-type-v q8_0 ^
  12. --jinja ^
  13. --no-mmap ^
  14. --mlock ^
  15. --main-gpu 1 --split-mode layer --tensor-split 54,46 ^
  16. --ctx-checkpoints 32 --checkpoint-min-step 1024 ^
  17. --parallel 1 --kv-unified ^
  18. --cache-ram 8192 ^
  19. --temp 0.6 --top-k 20 --top-p 0.95 --min-p 0.0 ^
  20. --repeat-penalty 1.0 ^
  21. --presence-penalty 0.0
Advertisement
Add Comment
Please, Sign In to add comment