Official implementation of the Msanii paper. The paper introduces a novel diffusion based model for audio synthesis. It outperforms existing diffusion models in terms of fidelity, long-term coherence and efficiency.

odewel (/əʊdwel/) is a library that can optimize machine learning models for better efficiency by loading weights only when needed for computation. This makes the models leaner and more nimble, and allows them to run on any hardware, regardless of its size or the hardware capabilities.

A tiny character level GPT implementation trained on shakespeare. The model leverages the recently introduced U-Net block from Msanii to capture both global and local dependencies between characters, resulting in text that has both short term and long-term coherence.