Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fine-Tuning Question? #13

Open
robot-gen opened this issue Jan 8, 2025 · 5 comments
Open

Fine-Tuning Question? #13

robot-gen opened this issue Jan 8, 2025 · 5 comments
Labels
question Further information is requested

Comments

@robot-gen
Copy link

Can I directly leverage the Cosmos-1.0-Diffusion-7B-Video2World checkpoint to continue fine-tuning it by using the training code from NeMo's Diffusion?

If not, where can I download the model checkpoint of the DiT5/7B model for NeMo fine-tuing?

Thanks!

@snowmanwwg
Copy link

Finetuning code will be available in the next day or so. Will let you know.

@ethanhe42
Copy link
Member

Hi @robot-gen , fine-tuning is available here https://github.com/NVIDIA/Cosmos/tree/main/cosmos1/models/diffusion/nemo/post_training#readme

@robot-gen
Copy link
Author

Thanks,this is a fantastic work! I am expecting the coming Video2World finetuning code~

@jpenningCA
Copy link

@robot-gen I'm a PM at NVIDIA for COSMOS, can you share your use case for Post-training/Fine-tuning using the Video2world WFM?

@mharrim mharrim added the question Further information is requested label Jan 24, 2025
@akshatd007
Copy link

@jpenningCA Hey, I would love to actually view the Video2World finetuning script. I used the 14B video2world on an A100 40GB. However, my use case for interior and lifestyle generations seem to be satisfied using the 7B varaint.

I wanted to fine tune it in order to reduce deformations and more importantly have the WFM model follow camera prompts more accurately. Thanks Joel

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
question Further information is requested
Projects
None yet
Development

No branches or pull requests

6 participants