RoMeo: Mitigating Dual-dimensional Outliers with Rotated Mixed Precision Quantization
Abstract
Mixed precision quantization has been adopted to accelerate large language models (LLMs) serving by leveraging high-throughput low-precision compute units in GPUs while preserving outliers in higher precision to maintain model accuracy. However, existing methods focus on mitigating single-dimensional channel-wise outliers, leading to model accuracy degradation when scaled to 4-bit precision.
Fields of Study & Subjects
Advanced Neural Network Applications
Numerical Methods and Algorithms
Low-power high-performance VLSI design
Publication details
- Journal / source
- ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming
- Pages
- 273–287
- Publication date
- 28 January 2026