THUDM

/

cogvlm2-llama3-caption

like 101

Follow

Z.ai & THUKEG 2.64k

Swapping import torchvision.transforms.functional_tensor as F_t in pytorchvideo to import torchvision.transforms.functional as F_t works fine for me.
I think that's from a change of torchvision lib.

That works! Thanks! @lly0571

Command to do the above patch in Colab. If you are not on Colab, change the path to your python location.
!sed -i '9s/.*/import torchvision.transforms.functional as F_t/' /usr/local/lib/python3.11/dist-packages/pytorchvideo/transforms/augmentations.py

After applying what @lly0571 suggested, the error is gone, but I encountered a new situation:
Error processing video A_turtle_is_walking_on_the_sand.MP4: 'CogVLMVideoForCausalLM' object has no attribute '_extract_past_from_model_output'