Defined in awkward.operations.convert on line 724.


Returns the names of the kernels library used by arrays. May be

  • "cpu" for libawkward-cpu-kernels.so;

  • "cuda" for libawkward-cuda-kernels.so;

  • "mixed" if any of the arrays have different labels within their structure or any arrays have different labels from each other;

  • None if the objects are not Awkward, NumPy, or CuPy arrays (e.g. Python numbers, booleans, strings).

Mixed arrays can’t be used in any operations, and two arrays on different devices can’t be used in the same operation.

To use "cuda", the package awkward-cuda-kernels be installed, either by

pip install awkward-cuda-kernels

or as an optional dependency with

pip install awkward[cuda] --upgrade

It is only available for Linux as a binary wheel, and only supports Nvidia GPUs (it is written in CUDA).

See ak.to_kernels.