About training code and scripts

[wytbwytb]

It's a nice work. When will you upload the training code and scripts ？

[SteveKGYang]

We mostly modified the scripts of FastChat (https://github.com/lm-sys/FastChat) for the fine-tuning process. You can look into that.

[NirmalManoj]

Hi @SteveKGYang, great work! Can you please release the code used for training bart-large and T5?

[NirmalManoj]

@SteveKGYang I want to fine-tune with bart-base, but with the same code, processing, etc. that your team used.

[biirving]

Why not just release it bro

[Zuhashaik]

Which fine-tuning method you guys used for this ?

• Unfreezing only last few layers
• LoRA
• Attached an extra decoder layer at last ?

[SteveKGYang]

@Zuhashaik We used full fine-tuning, which means all parameters are tuned.