Repetition in model

[jxnding]

Using the same Q4_K_XL I don't have these issues with llamacpp but i do with lucebox.

DFLASH_FP_USE_BSA=1 DFLASH_FP_ALPHA=0.85
python scripts/server.py
--target ~/models/Qwen3.6-27B-UD-Q4_K_XL.gguf
--draft ~/models/draft/dflash-draft-3.6-q8_0.gguf
--ddtree-budget 40
--port 8080
--max-ctx 200192

[davide221]

Hey @jxnding thanks for reporting it! Do you get this for any prompt?