torch-spyre

This repository provides architectural documentation and backend support specifically for the IBM Spyre hardware accelerator. It enables the mapping of neural network operations onto the Spyre silicon architecture to facilitate high-performance compute tasks.

The target audience includes hardware engineers, systems architects, and machine learning researchers working on custom silicon integration. It is designed for professionals needing to understand the interface between PyTorch-based models and the IBM Spyre physical compute layer.

Users require a functional PyTorch environment and access to the IBM Spyre hardware ecosystem. Familiarity with low-level hardware acceleration backends and custom operator implementation within the PyTorch framework is necessary to effectively leverage the provided architectural documentation.