So is it instruct or base model?
Will it follow instructions or just keep going on?
The Mistral 7B is definitely the Instruct. The CatPPT has this section in the model's card: https://huggingface.co/rishiraj/CatPPT-base#inference-procedure
But it seems to be the base model now that I look at it! (the author mentioned this in the discussion!)
CatPPT seems to be a merge of two other models, but not more information about them is available. My script must have picked it up as a trending Mistral-based model to merge it. However, it does work, even the quantized on CPUs. Not sure how much different it would be compare to Mistral itself:
The Mistral 7B is definitely the Instruct. The CatPPT has this section in the model's card: https://huggingface.co/rishiraj/CatPPT-base#inference-procedure
Do you know how did the merge happened? If head was from mistral instruct then it most probably is an instruct model, however, if it is following your instructions but is not an instruct model then it is pretty damn good.
Can you confirm if it can read text (paragraphs, or large documents) and can create json regarding that data? For example a short story it could return what story is about, number of characters, main characters, etc in json.
And for what prompts and prompt format it works the best.
Thank you again for your contributions.
Yes, it does:
{
"arguments": {"title": "My Weekend To-Do List", "tasks": ["Clean room", "Wash clothes", "Cook dinner", "Buy groceries"]},
"name": "create_todo_list"
}
The instructions can be part of your SYSTEM PROMPT or part of the prompt itself:
Yes, it does:
{ "arguments": {"title": "My Weekend To-Do List", "tasks": ["Clean room", "Wash clothes", "Cook dinner", "Buy groceries"]}, "name": "create_todo_list" }The instructions can be part of your SYSTEM PROMPT or part of the prompt itself:
Neat. Thanks for the model.