RoMeo: Mitigating Dual-dimensional Outliers with Rotated Mixed Precision Quantization

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.