Naming scheme
#1
by
Vezora
- opened
Out of curiosity why was the naming scheme “base-7b-v0.2” when the mistral v0.1 was used for continued pre-training?
Either way congratulations on the model, this is seriously awesome. I love it! And thank you for apache 2.0! ❤️🤗👏
Hello,
Our naming scheme probably isn't the best, we actually had a v0.1 that was finetuned using another format of benchmarks. Since lm-eval-harness has since then implemented the benchmarks natively we had to finetune it again with the expected format.
Thank you very much for your kind words!
Oh I understand, that makes sense! Once again, thank you, y'all are so awesome for this model! ❤️
Vezora
changed discussion status to
closed