I have difficulty to trigger thinking process
same.
Some issue with the config and conversion into GGUF?
The thinking should be wrapped in and , let me know if you are able to find these tokens in the LLM generated outputs!
The thinking should be wrapped in and , let me know if you are able to find these tokens in the LLM generated outputs!
sometimes it would just output token without thinking.
maybe it can use more training?
The thinking should be wrapped in and , let me know if you are able to find these tokens in the LLM generated outputs!
sometimes it would just output token without thinking.
maybe it can use more training?
On larger models, I usually always find a way, using one of those:
- "Match your effort to the task. And when it gets tough, take as long as you need to think before you start answering."
or - Something like "The goal is to find []" or "The goal is to figure out []"
@michaelzhiluo Thanks for your help, but there have been a formatting issue with your reply, can you share again? :)