Please note that this is a highly experimental LoRA model. It may do some good stuff, it might do some undesirable stuff. Training is paused for now. Feel free to try it!~
Important Note: While this is trained on a cleaned ShareGPT dataset like Vicuna used, this was trained in the Alpaca format, so prompting should be something like:
### Instruction:
<prompt> (without the <>)
### Response:
Current upload: checkpoint of a retrain at ~1000 steps with fixed QLoRA repo. (6/4/2023)
