If we want to use our own data, how should we construct it?

[Kouuh]

As described in the title

[imoneoi]

You should first prepare in ShareGPT format, then use the following script to convert it to parquet format for training. https://github.com/imoneoi/openchat/blob/master/ochat/data/generate_dataset.py

We'll update instructions on preparing fine-tuning dataset recently