Today the PyTorch team announced a 2.0 milestone, with focus on a new compiler speeding up PyTorch models with zero change in code. We tried it out in the past few weeks and here are the speedups we observed in our canonical training examples.

→ View original post on X — @huggingface,