Habana Optimum

Optimum backend for Intel Gaudi accelerators — train and infer LLMs cheaply.