Questions on Training Dataset

[BFMarks]

Thanks for the great repo, quick questions though:

• What is the minimum amount of data rows would you recommend for a new dataset?
• What is the exact ideal format for reformatting a new dataset?
• Can it not be that negative with a different dataset (and lighten up a little)?

[pender]

What is the minimum amount of data rows would you recommend for a new dataset?

I'd say a few megabytes compressed is probably the minimum. Bigger models will need greater amounts of data to avoid overfitting.

What is the exact ideal format for reformatting a new dataset?

Multiple lines of dialogue, each starting with "> " (note the space) and an empty newline between dialogues.

Can it not be that negative with a different dataset (and lighten up a little)?

Yes, a data set with less negative content will probably result in a bot with a less negative tone.