how to disable <think> with llama.cpp
#9
by
bobchenyx
- opened
I'm having this trouble with running llama-bench.
Not getting any error / results.
I'm wondering if it's due to this stuff. and would like to ask how to disable this property.
just put "nothink" in the system prompt
just put "nothink" in the system prompt
do you mean I should modify some .gguf before running llama-bench?
Your problem is probably this qwen3_unsloth_dynamic_ggufs_128k_context_bug_fixes (the chat_template was incompatible with llama.cpp, and falling back to chatml.
I'm guessing llama-bench is stricter and will fail rather than fail silently fallback given it's supposed to be reproducible.
Looks like they fixed it just after you posted, so try re-downloading the .gguf now