Closed matbee-eth closed 1 month ago
That said, we will definitely try to support tuning vision modules as well.
I think one option is to wrap the vision modules with HF PEFT to support e.g. low-rank updates to the vision module. See xtuner implementation: https://github.com/InternLM/xtuner/blob/main/xtuner/model/llava.py
Yes that is the idea. Thank you for sharing your thoughts @fedshyvana
Supported with #14. In the example scripts there are added arguments like TRAIN_VISION_ENCODER
, USE_VISION_LORA
, TRAIN_VISION_PROJECTOR
. Feel free to try it out.
Curious why you made that decision?