I went for Triton because it was already closely integrated into PyTorch. Taichi looks really polished in comparison, but not tried it yet!

→ View original post on X — @alexjc