Comment by echelon

> No Python runtime, no PyTorch, no CUDA toolkit required at inference time.

This is amazing, Salvatore! Please spend some more time here and free us from the CUDA toolkit and Python.