Quantization and Calibration

This document serves as an overview of the quantization and calibration group. To avoid mixing traditional models with foundation models, the content has been split into the following two documents:

• Traditional Model Quantization and Calibration
• Foundation Model Quantization and Compression

Content Boundaries

• Traditional models: PyTorch/ONNX/MindSpore post-training quantization and quantization-aware training.
• Foundation models: low-memory quantization, hybrid calibration datasets, and FA3 quantization.