Openchat3.5 training data

[jzhang38]

Congrats for the V3.5 release! May I ask if there are plans to release your finetuning data, just like what you have been always doing with your previous release?

[imoneoi]

Thanks! Sorry the dataset contains a bit private data, but we are considering releasing the open subsets of the dataset.

[yucc-leon]

Similar question. Did V3.5 still use the same strategy as the paper openchat describe? Or there were other changes and you will publish a new-version report?

[imoneoi]

Similar question. Did V3.5 still use the same strategy as the paper openchat describe? Or there were other changes and you will publish a new-version report?

Yes, it's mainly based on the method that the paper described (except with more data). We may publish a new report if new modifications are made.

[timothylimyl]

@imoneoi if all data is generated using openai api, where does the private data come from?

[mietekrmd]

@imoneoi if all data is generated using openai api, where does the private data come from?

I quess from books

[timothylimyl]

does your dataset have system prompts?