v1.0: Open-omics-alphafold release

• A PyTorch implementation of AlphaFold2 (v.2.2.2) monomer accelerated using 4th generation Intel® Xeon® CPU.
• Accelerated gating attention and triangle multiplication using Intel’s hardware and software support for deep learning
  • Using software support: Developed efficient version of the modules using Intel’s Tensor Processing Primitives (TPP) (https://github.com/libxsmm/tpp-pytorch-extension/tree/main/src/csrc/alphafold)
  • Using hardware support: Applied Bfloat16 quantization to use Intel® Advanced Matrix Extensions (AMX), the built-in hardware support for deep learning on 4th generation Intel® Xeon® CPUs.
    Can perform model inference on proteins of length up to ~9k residues on a 1TB memory machine.
• More than 10x faster in the model inference stage over prior CPU implementation.
• This is being released along with hh-suite and hmmer accelerated using AVX-512 for faster preprocessing.