Hi, I use ColabFold on the web and suddenly it causes issues. MSA appears as completed and green but shows errors and when passed onto the Alphafold it doesn’t work.
Unable to initialize backend ‘cuda’:
FAILED_PRECONDITION: No visible GPU devices.
Unable to initialize backend ‘rocm’:
NOT_FOUND: Could not find registered pl.
I also tried re-running the same protein, which I successfully did in December and it gives me the same error.
I was trying to find the information about this, but I was unsuccessful. I would appreciate any insights, thanks.
We just updated the server to the (pending) next release. This may have caused some interruptions, and there may be a few problems uncovered! This is the broadcast notice we have now posted on the UseGalaxy.org server.
This is the server where you are working at, correct? I’m going to ask our administrators to take a look in general, but if you could share the details for this specific error that might help, too. I would also say to try a rerun again now since yesterday was when we actually updated the software, and maybe that improved the prior problem or some connection was lost temporarily and is now back (ColabFold is a bit special/complicated).
Would you be able to share a link to a small history with the failure? You could copy just the inputs and outputs into a new history and share that slice. You could even start your rerun in that same new history and we can watch it, too. How to share is in the banner of this forum, also here. → How to get faster help with your question. The link can be posted back as a reply.
Thanks for letting us know about the error and hopefully a rerun is enough but we definitely want to follow up!
This is what I found. I hope this help, if there is other information you need me to access or if you want this in a different format please guide me of how to do so. I am sorry I am quite new to this.
This was actually the new attempt on colabfold MSA and the colabfold Alophafold is from Friday as the reattempt is running just now. Since MSA error is the same as previously I expect the same issue with Alphafold, so I just wanted to share the output sooner. If it will come out different I will make an update.
Thanks for sharing the details! I’m not sure that first run actually failed.
The warning messages about the directory paths can be ignored – the tool is running in a container environment so it sort of falls through the default locations then into the one we want it to use, and is chatty as it does that. We can’t suppress those messages, so just know you can ignore those.
But the second job is interesting and does look like it failed, so maybe some clues are in there. If you could show the job info page for that red dataset 144 plus any messages in the dataset view itself, that would be great. If you could also send in a bug report from that red dataset, we can look closer at it. Consider putting a the link to this forum topic in the bug report comments so we can link the two together easier.
