request for gradio UI , dataset captioning scripts

[GeeveGeorge]

Feature request / 功能建议

@a-r-r-o-w @sayakpaul

it would be great if a gradio UI could be made for fine-tuning CogVideoX.
Users can provide path to videos (no need to upload) , just a path.
Then hit fine-tune lora (it needs to split videos, caption it (say using BLIP), and prepare the dataset for fine-tuning.
Then run fine-tuning on single GPU and get the lora.
Users can also test inference on a different gradio UI by specifying the lora path.

It would be great if you could take this into consideration to make it super user friendly.

I could help as well making the gradio UI, if the data prep and captioning scripts are available.

Motivation / 动机

better UX UI for new comers.

Your contribution / 您的贡献

gradio UI

[sayakpaul]

The repository is supposed to be hackable so, a UI is currently not on our priority. Feel free work on separately and we will try to help. Regarding captioning, see #34.

[sayakpaul]

With the existing scripts, it should not be too hard. Feel free to start one if you want. Multi-frame, multi-aspect sampling are already supported. Please read dataset.md doc.

[a-r-r-o-w]

I think it might be interesting and make things easier to work with, considering that all users may not be researchers. I did get a few DMs requesting a UI too, as the different configurations can be hard to understand.

@GeeveGeorge As Sayak mentioned, we are short on time so cannot prioritize, but if you could create a prototype and open a PR, we would be more than happy to help assist. Thanks!