train error

[LeeKeyu]

Hi authors, thanks for your impressive work.
I tried to implement training using bash accelerate_train.sh, and encountered this error:

Traceback (most recent call last):
File "/RealCam-I2V/finetune/train.py", line 21, in
main()
File "/RealCam-I2V/finetune/train.py", line 15, in main
trainer_cls = get_model_cls(args.model_name, args.training_type)
File "/RealCam-I2V/finetune/models/utils.py", line 48, in get_model_cls
raise ValueError(f"Model '{model_type}' is not supported")
ValueError: Model 'cogvideox1.5-i2v' is not supported

I checked utils.py and found that SUPPORT_MODELS is empty. Are there any bugs with this training config? How should I modify the training code? Thank you a lot.

[kelisiyaer-ctrl]

How did you solve the environment issue? When I use torchcodec==0.0.2, I encounter the error "cannot import name 'VideoDecoder' from 'torchcodec.decoders'". Also, would you mind sharing your CUDA and PyTorch versions? Thank you.

[Weather-etc]

How did you solve the environment issue? When I use torchcodec==0.0.2, I encounter the error "cannot import name 'VideoDecoder' from 'torchcodec.decoders'". Also, would you mind sharing your CUDA and PyTorch versions? Thank you.

VideoDecoder is introduced to torchcodec in version 0.1, so you can try torchcodec==0.1 instead. Refer to torchcodec0.1 doc

[MuteApo]

Hi authors, thanks for your impressive work. I tried to implement training using bash accelerate_train.sh, and encountered this error:

Traceback (most recent call last):
File "/RealCam-I2V/finetune/train.py", line 21, in
main()
File "/RealCam-I2V/finetune/train.py", line 15, in main
trainer_cls = get_model_cls(args.model_name, args.training_type)
File "/RealCam-I2V/finetune/models/utils.py", line 48, in get_model_cls
raise ValueError(f"Model '{model_type}' is not supported")
ValueError: Model 'cogvideox1.5-i2v' is not supported

I checked utils.py and found that SUPPORT_MODELS is empty. Are there any bugs with this training config? How should I modify the training code? Thank you a lot.

Thanks for catching this. I missed these files because of the dynamic model registration in __init__.py. They've now been added in the latest commit.

[MuteApo]

How did you solve the environment issue? When I use torchcodec==0.0.2, I encounter the error "cannot import name 'VideoDecoder' from 'torchcodec.decoders'". Also, would you mind sharing your CUDA and PyTorch versions? Thank you.

Apologies for the confusion, that was a typo on our part. The correct version for torchcodec is 0.2.0, which is required for compatibility with torch 2.6.