jiaor17 / diffcsp-pp Goto Github PK

Thank you for open sourcing your code！
However, I notice that DiffCSP-PP does not provide prediction code. Can I use DiffCSP-PP trained weight files for prediction using DiffCSP code?
Or will you open source the prediction code for DiffCSP-PP in the future？

Hello，In the DiffCSP-PP-main/diffcsp/pl_modules/diffusion.py , within the def sample module, what is the role of diff_ratio, and has it always been 1?

Dear team,

I am trying to replicate the results by using the same training script for dataset=perov_5 following instructions here.
When loading the model in a Jupyter notebook, I get the following error message:

RuntimeError: Error(s) in loading state_dict for CSPDiffusion:
size mismatch for decoder.csp_layer_0.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.csp_layer_1.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.csp_layer_2.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.csp_layer_3.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.csp_layer_4.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.csp_layer_5.edge_mlp.0.weight: copying a param with shape torch.Size([512, 1798]) from checkpoint, the shape in current model is torch.Size([512, 1801]).
size mismatch for decoder.lattice_out.weight: copying a param with shape torch.Size([6, 512]) from checkpoint, the shape in current model is torch.Size([9, 512]).

Hi , would you please share a complete list of requirement.txt, i find the installation is very difficult now .