Fine-tuning

Give two keys and get a model tuned on your data, in your own repository, with a report of what the tuning changed in the model's behaviour. The training runs on your GPU account; your keys are stored encrypted and deleted when the GPU part ends.

New job

Data repository (Hugging Face dataset)

CSV, JSONL or Parquet with a prompt column and a reply column.

Output repository (Hugging Face model)

Created private if it does not exist. The tuned weights land here.

Base model

Hugging Face token

Needs read on the data repository and write on the output repository.

Modal token ID

Modal token secret

From modal.com, Settings, API tokens. The training uses one GPU on your account (L4, else A10G, else T4).

Training rows

Epochs

Fine-tuning

New job

Your jobs