radi-cho / datasetgpt Goto Github PK

From what I've understood, it should be :
initial_utterances: List[str] = field(default_factory=lambda: ["Hello."])
Previously it was:
initial_utterances: List[str] = "Hello."

When I was trying to generate conversations with the previous version, I found the model gave me six conversations even when I set num_samples to 1.
Then I discovered that the "initial_utterance" of each conversations turned out to be "H", "e", "l", "l", "o". It does not make sense to let a single letter be the initial utterance of a converstaion.

Is it possible to adding this project to multilanguage support? I want to finetune Llama with my local language but there is not any dataset for this

It would be good if the community would curate one large dataset for people to share and contribute to together.

Currently, the initial message given to agent1 in the conversation generation mode is not being formatted. For consistency with the rest of the functionality, it would be better to apply a template to it so options can be shared between agent1, agent2, and initial_utternace.