alphafold: Getting "ValueError: Cannot create a tensor proto whose content is larger than 2GB." for smaller sequences

Hello! I’m getting the error “ValueError: Cannot create a tensor proto whose content is larger than 2GB.” when running AlphaFold jobs for proteins longer than ~650 residues. I have tried using --max_template_data=1900-01-01 in order to limit MSA size, but this has not helped. I using 1 GPU, 8 CPU cores, and 150GB memory on my university’s supercomputer. I am not interested in using --preset=reduced_dbs. Thanks!!

About this issue

Original URL
State: closed
Created 3 years ago
Comments: 26 (1 by maintainers)

Most upvoted comments

@hegelab - there are 5 commits on your main branch, one of which even creates a log file at a hardcoded path that won’t exist on most machines. This goes way beyond the 2-line patch being advocated above.

I’d suggest creating a separate branch with the just the changes (without renaming/moving to .orig) you think are important, and create a PR. That way, DeepMind can incorporate or comment on your changes.

chrisroat on Oct 25, 2021

Hi, would you mind sharing your input sequence and/or the MSA size? We wouldn’t expect to see this error from inputs of that length but there have been a few other reports like this, so we would like to investigate it further.

Here is a link to a Google Drive folder containing the .fasta file and .a3m file for my AlphaFold job: https://drive.google.com/drive/folders/1bGS80lNFlo9xUDXQs71K5im_1ZihBPut?usp=sharing

If there are any more files or contextual information that I should provide, please let me know! Thanks so much for taking a look at my issue.

DavidB256 on Jul 30, 2021