Re-Finetune llama2
ismailousa opened this issue · 0 comments
ismailousa commented
Remarks
Initial checkpoints generates a lot of random text to be considered proper code. This issue aims to address it
Requirements
- Update signature generator to match Adaptor.d.ts format
- Add parameter and return objects to signature JSDoc
- Remove non rest api functions from dataset
- Multiply data (too little data initially)
- run peft then supervised finetuning