Accelerating ProtST on Intel Gaudi 2 — 1.76x Inference Speed

AI Impact Summary

Intel has accelerated Protein Language Model ProtST on the Intel Gaudi 2 accelerator, achieving competitive performance against NVIDIA A100 hardware. This represents a significant speedup in both inference (1.76x) and fine-tuning (2.92x), making ProtST more accessible for researchers and developers. The use of the Optimum for Intel Gaudi library allows for straightforward porting of existing transformer-based scripts, facilitating rapid experimentation and deployment.