issue about checkpoints

[LouisDRR]

Hello author, thank you for open-sourcing this work. However, I encountered an issue while trying to run the project: even though I have already downloaded the 2B_pretrained model weights from nvidia/DreamDojo, why does the system still require me to download various model weights from nvidia/Cosmos-Predict2.5-2Bwhen I run eval.sh?

[jiduster]

I've also encountered the same issue. I think the problem is mainly due to the provided huggingface revision id for Cosmos-Predict2.5-2B defined in 'DreamDojo/cosmos_predict2/_src/imaginaire/utils/checkpoint_db.py' does not exist. I managed to resolve it by changing the revision numbers to an existing revision id. I think you can replace all the ""6787e176dce74a101d922174a95dba29fa5f0c55"" in checkpoint_db.py with the exact revision id that you observe in your local huggingface hub snapshot directory.

But still, this is only a temporary solution to the issue. For a complete remedy, maybe we should wait for official answers from the authors.

[LouisDRR]

I've also encountered the same issue. I think the problem is mainly due to the provided huggingface revision id for Cosmos-Predict2.5-2B defined in 'DreamDojo/cosmos_predict2/_src/imaginaire/utils/checkpoint_db.py' does not exist. I managed to resolve it by changing the revision numbers to an existing revision id. I think you can replace all the ""6787e176dce74a101d922174a95dba29fa5f0c55"" in checkpoint_db.py with the exact revision id that you observe in your local huggingface hub snapshot directory.

But still, this is only a temporary solution to the issue. For a complete remedy, maybe we should wait for official answers from the authors.

Now I understand why I need to download cosmos-reason1 and Wan2.1 VAE. The NVIDIA team mentioned in the paper of cosmos-predict2.5 that the model uses cosmos-reason1 as a text encoder and Wan2.1 VAE as a visual tokenizer. But I still don’t understand why I need to download the pre-trained models of both cosmos-predict2.5-2B and cosmos-predict2.5-14B