How to finetune this model?
#1
by
Labmem009
- opened
Interesting work!
Could u pls offer the finetune code or script, so I can use this jp-VLM in downstream tasks?
Thanks a lot!
This first improves the FullFinetuning of the language part using the very common DPOTrainer.
This is trl's DPOTrainer.
BETA was done with 0.1 and all we need to do is create a dataset, Chosen and Rejected, and use the dataset with Score.
After that, the images were trained using STIC.
https://github.com/yihedeng9/STIC
Reference.