openfold
openfold copied to clipboard
Two incorrectly imported weights in multimer-branch
When running the multimer branch with the inference-example command in the readme, I get the following error
File "/root/openfold/openfold/utils/import_weights.py", line 630, in import_jax_weights_ assert len(incorrect) == 0 AssertionError
printing the 'incorrect' list, I get these two entries: ['alphafold/alphafold_iteration/structure_module/fold_iteration/invariant_point_attention/q_scalar_projection//weights', 'alphafold/alphafold_iteration/structure_module/fold_iteration/invariant_point_attention/q_scalar_projection//bias']
If there is any information I could provide to fix this problem, please let me know. Note: I had some installation issues in my environment and ended up using python 3.8 instead of 3.7, numpy 1.22.1 instead of 1.21.1, torch 1.11 instead of 1.10 and cudatoolkit 11.7 instead of 10.2.
PS: For the import AA-chains, I use colon (':')-delimited fasta files, the same way colabfold requires its input formats. But I guess, this is not influencing the weight-loading anyways.