Closed
Description
Hello,
I have a dataset in .jsonl format of the following format:
{"messages": [{"role": "system", "content": "some system msg"}, {"role": "user", "content": "some user input"}, {"role": "assistant", "content": "some expected output"}]}.
I see there is an option to specify conversation_style=openai and source=json in the config files. However the tutorial for Llama3 finetuning did not mention this.
I was able to get finetuning running, but the result wasn't very good and I was just wondering if my workflow is correct at all, (also I couldn't seem to input the system prompt in inference testing, any ideas?)
Thank you in advance
Metadata
Metadata
Assignees
Labels
No labels