Brevitas Quantization Library - Pablo Monteagudo Lago, AMD

This talk covers the Brevitas Quantization Library, an open-source PyTorch resource developed by AMD that facilitates research into advanced quantization methods such as Qronos and MixQuant. The speaker discusses how Brevitas supports post-training quantization (PTQ), enabling practitioners to effectively use a unified environment for modern PTQ algorithms like SpinQuant and AutoRound without the need for retraining. By utilizing the latest features in PyTorch, Brevitas allows for innovative experimentation, demonstrating how its modular components can lead to new avenues in quantization research and applications. The session provides insights on employing Brevitas for a comprehensive quantization workflow, highlighting its flexibility and integration with other frameworks.