<ul><li>AWS Neuron SDK for developing and running new kernels on AWS Trainium and AWS Inferentia.</li><li>The NKI interface joins another API that enables NeuronCore-v2 programmability.</li><li>By exposing APIs for Neuron kernel customization, the SDK empowers developers to create and/or optimize the low-level operations, greatly increasing the opportunity for running ML workloads on Trainium and Inferentia.</li><li>The NKI documentation includes a dedicated section on the architecture design of NeuronCore-v2 and its implications on custom operator development.</li><li>Similar to other dedicated AI chips, NeuronCore-v2 includes several internal acceleration engines, each of which specializes in performing certain types of computations.</li><li>The NKI API Reference Manual details the Python API for kernel development.</li><li>The second method for creating a custom Neuron kernel involves building a C++ operator for the GpSimd engine.</li><li>Through its high-level Python interface, the NKI APIs expose the power of the Neuron acceleration engines to ML developers in an accessible and user-friendly manner.</li></ul>

On the Programmability of AWS Trainium and Inferentia

Discover more