NiiCole

/

vivit-b-16x2-kinetics400-finetuned-lora-ucf101-40-50prms-t2

like 0

you can use the same lora config as the vit implementation, the target modules should also match the one in your architecture . It works pretty wel
l for fine tuning

Thanks for the feedback!! I'm pretty new to vision models, as I'm coming from LLMs. Do you have documentation for this? Or where can i find how this is done? I saw a post on hf about vit lora, but not sure if that's what you meant. Thanks again for the help

yeah, i usually work on vision models . This should give you a clear idea
https://huggingface.co/docs/peft/task_guides/image_classification_lora

yeah, i usually work on vision models . This should give you a clear idea
https://huggingface.co/docs/peft/task_guides/image_classification_lora

Thanks!