Presents ZipCal, a model-agnostic calibration data selection strategy for pruning and quantization that is 240x faster than model-based methods.
March 18, 2026
Original Paper
Frequency Matters: Fast Model-Agnostic Data Curation for Pruning and Quantization
arXiv · 2603.16105
The Takeaway
Selection of calibration data is a major bottleneck in model compression; by using Zipfian power laws to maximize lexical diversity, this method achieves state-of-the-art results without expensive perplexity calculations.
From the abstract
Post-training model compression is essential for enhancing the portability of Large Language Models (LLMs) while preserving their performance. While several compression approaches have been proposed, less emphasis has been placed on selecting the most suitable set of data (the so-called \emph{calibration data}) for finding the compressed model configuration. The choice of calibration data is a critical step in preserving model capabilities both intra- and inter-tasks. In this work, we address th