Problem Description

I want to fine tune Video-LLaVA to make it uncensored

Here's the model in a hosted playground: https://replicate.com/nateraw/video-llava

Link to finetune lora script:
https://github.com/PKU-YuanGroup/Video-LLaVA/blob/main/scripts/v1_5/finetune_lora.sh

And the existing video fine tuning dataset:
https://github.com/PKU-YuanGroup/Video-LLaVA/blob/main/TRAIN_AND_VALIDATE.md

Acceptance Criteria

Communicate with me about the approach. Then:
Set up a fine tuning repo for Video-LLaVa
Model:
https://github.com/PKU-YuanGroup/Video-LLaVA

here's using it in transformers:
https://github.com/PKU-YuanGroup/Video-LLaVA/issues/156

Ideally in python
I want to modify the existing fine-tune dataset & provide additional examples
It should be set up such that:
I can import my dataset & train this on a cloud provider (e.g. modal labs, google collab)
I can use a script to deploy it to a host

If this works out I can hire you to fine tune llama3-llava-next image to text model as well