I’m helping out with a community research project trying to preserve and study the behavior of older GPT-3.5-Turbo models, particularly 0613 and 16k.<\/p>\n
I’m trying to find any existing data from when these models were available, especially:<\/p>\n
1)prompt\/response datasets or cached model outputs<\/p>\n
2)old benchmark runs with individual predictions<\/p>\n
3)API logs containing model responses<\/p>\n
4)token-level logprobs or top-token alternatives, if available<\/p>\n
5)public datasets, repositories, or papers that might point us to any of the above<\/p>\n
Even a small collection would be useful! We’re particularly interested in data where the model version and collection method are documented, so we can distinguish verified historical outputs from data collected through other providers or newer models.<\/p>\n
If you’ve saved anything like this, or know of a repository or person I should look into, I’d really appreciate a pointer.<\/p>\n
Thanks :-))<\/p>\n<\/div>\n
submitted by \/u\/New_Today172 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
