RFdiffusion Colab Setup Cell Not Working: ModuleNotFoundError: torchdata.datapipes

[Ehsanbm96]

Hi RFdiffusion Team,

I am encountering an issue while trying to run the RFdiffusion Google Colab notebook. I am using the latest version of the notebook. When running the setup cell, I receive a ModuleNotFoundError: No module named 'torchdata.datapipes'. The traceback indicates this error occurs within the dgl library, which seems to require torchdata.datapipes. I have attempted to resolve this by explicitly installing dgl (version 2.1.0+cu121) and torchdata using pip install commands. While the installations complete successfully, the ModuleNotFoundError for torchdata.datapipes persists when the code proceeds to imports.

It appears there might be a dependency conflict or a specific compatibility requirement between dgl and torchdata in the Colab environment. Could you please advise on how to resolve this issue? Any guidance on compatible versions or necessary environment configurations would be greatly appreciated.

Thank you for your help!

error: "ModuleNotFoundError Traceback (most recent call last) in

/content/RFdiffusion/inference/utils.py in 6 import torch 7 import torch.nn.functional as nn ----> 8 from diffusion import get_beta_schedule 9 from scipy.spatial.transform import Rotation as scipy_R 10 from util import rigid_from_3_points

8 frames /usr/local/lib/python3.12/dist-packages/dgl/graphbolt/base.py in 6 import torch 7 from torch.utils.data import functional_datapipe ----> 8 from torchdata.datapipes.iter import IterDataPipe 9 10 from ..utils import recursive_apply

ModuleNotFoundError: No module named 'torchdata.datapipes' "

[rclune]

Is this link how you are accessing the notebook? https://colab.research.google.com/github/sokrypton/ColabDesign/blob/main/rf/examples/diffusion.ipynb

Running what is already present in the notebook does not produce this error, what settings did you use?

If you are using a system with an AMD architecture, you can also use the Docker image to run RFdiffusion. There's more information about Docker images here: https://sites.google.com/omsf.io/rfdiffusion/getting-started/docker-images