olive-arena: automates model optimization for HuggingFace models using Microsoft Olive
olive-arena automates the optimization of HuggingFace models using Microsoft Olive, producing ONNX models and measuring performance metrics like perplexity and tokens/sec. It's designed for use with external coding agents to iteratively improve model performance. The project includes a leaderboard to track experiments and identify Pareto-optimal configurations.
Visit author’s GitHub →icnatspell/olive-arena