AlphaFold v3.0.4

Version 3.0.4 makes AlphaFold 3 runnable on CPU-only machines and on Apple Silicon. Moreover, there are some performance improvements and small bug fixes. We thank everyone who reported issues, proposed new features, and submitted PRs!

Change log

Below is a selection of new features, performance improvements and bug fixes. Full change log: v3.0.3...v3.0.4

New features

• AlphaFold 3 now runs on CPU-only machines! It is slow without a GPU, but you can run it anywhere. Thanks @wuooo339 for discovering the fix to make this work.
• AlphaFold 3 now runs on Apple Silicon (and can use the GPU to accelerate the inference time). Thanks @SuhasSrinivasan.
• Enable Google Cloud Storage (gs://) path support in AlphaFold3 using etils.epath.
• Added chain IDs in the summary confidence JSON for easier use. Thanks @ntnn19.

Performance improvements

• Reduce memory use for the outer product mean by replacing 2 einsums with a single one. Thanks @mooreneural.
• Updated JAX 0.9.1 -> 0.10.2 and Tokamax 0.0.11 -> 0.0.12, which should significantly speed up inference times and fix unified memory support on Blackwell GPUs.

Bug fixes

• Early validation that RNG seeds are uint32s. Thanks @ntnn19.

Documentation

• Add a note that user-provided CCD can be used for custom modified polymer residues. Thanks @Fred-Jing.

We thank the following people for their pull requests and/or great suggestions: @Fred-Jing, @mooreneural, @ntnn19, @SuhasSrinivasan, @wuooo339.