This page lists known limitations.

Triton Related Limitations

1. triton.compile()

To use triton.compile(), one needs to explicitly use triton.compile(kernel, device_type='xpu') instead of just use triton.compile(kernel).

By default, triton.compile() uses CUDA instead of XPU. This will lead to errors.