Closed teknium1 closed 1 year ago
So, I was cleaning up the bucket and accidentally deleted it 🤦
I updated the readme to only link to the valid huggingface datasources, including a new one that is all GPT-4 (plus a few thousand simple math prompts I was using, to test if it helps prevent the model from degrading its math abilities during fine tuning)
Feel free to use those for now if you wish; I'll see if I can recover the original.
The smaller GPT4 subset fine-tuned model (not publised yet) seems to perform better anyways.
So, I was cleaning up the bucket and accidentally deleted it 🤦
I updated the readme to only link to the valid huggingface datasources, including a new one that is all GPT-4 (plus a few thousand simple math prompts I was using, to test if it helps prevent the model from degrading its math abilities during fine tuning)
Feel free to use those for now if you wish; I'll see if I can recover the original.
The smaller GPT4 subset fine-tuned model (not publised yet) seems to perform better anyways.
Beautiful, appreciated!
When I click the 3.5-turbo dataset link it does this: