Yu-won Lee
Yu-won Lee
@JasonLeeUT @7998857 I'll upload my docker image in the dockerhub. I was using cuda 12.4 (via docker) and updated the flash-attn into the lastest one right now. Maybe there are...
I've uploaded a pre-build image to simplfying the setup.
I've uploaded the docker image and how to use it in the readme. Also mentioned about the system env. Thanks for debugging!
Sorry, I haven't seen the error. I'll check what is wrong with it. Thanks for letting me know.
``` [ { "id": "000000033471", "image": "000000033471.jpg", "conversations": [ { "from": "human", "value": "\nWhat are the colors of the bus in the image?" }, { "from": "gpt", "value": "The bus...
@Gopi-Uppari Thanks for the interest in the project! Your expirience would be very valuable for developing the project. Feedbacks and issues are always welcome!
Yes you could, but you need to fix a the code a little bit, in `data.py` and `parametrs.py`.
The model does support batch however, the processor dosen't so you should make a function for handling bathc inference. For example it is my code when using litserve. ``` def...
Can you make this PR merged? This would be helpful for pepole who wants to finetune the model, following some other papers like LLaVA-Next. By my experience, setting the learning...
yes it's becuase of lora layers are only applied to LLMs and the non-lora layers should be saved too (such as vision encoder).