mistralai/mistral-finetune for classification task

#175
by mikali - opened

I'm interested in using the Mistral fine-tuning package found here: https://github.com/mistralai/mistral-finetune/tree/main

My dataset is in CSV format with two columns: "text" (containing the text to be classified) and "label" (an integer between 0 and 3).

The documentation mentions two data file types for pre-training:

  • pretrain: {"text": "Text contained in document n°1"}
  • instruct: { "messages": [ { "role": "user", "content": "User interaction n°1 contained in document n°1" }...
    The pretrain format seems irrelevant for my classification task, and the instruct format doesn't seem to fit my labeled data structure.

How can I adapt my CSV data to be compatible with this fine-tuning package for a classification task? Are there any examples or specific configurations I should be aware of?

mikali changed discussion status to closed

Sign up or log in to comment