Rotate, Clip, and Partition: Towards W2A4KV4 Quantization by Integrating Rotation and Learnable Non-uniform Quantizer

Documenti analoghi

• Grouped Sequency-arranged Rotation: Optimizing Rotation Transformation for Quantization for Free
  di: Choi, Euntae, et al.
  Pubblicazione: (2025)
• NSNQuant: A Double Normalization Approach for Calibration-Free Low-Bit Vector Quantization of KV Cache
  di: Son, Donghyun, et al.
  Pubblicazione: (2025)
• LAQuant: A Simple Overhead-free Large Reasoning Model Quantization by Layer-wise Lookahead Loss
  di: Choi, Euntae, et al.
  Pubblicazione: (2026)
• RotateKV: Accurate and Robust 2-Bit KV Cache Quantization for LLMs via Outlier-Aware Adaptive Rotations
  di: Su, Zunhai, et al.
  Pubblicazione: (2025)
• QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving
  di: Lin, Yujun, et al.
  Pubblicazione: (2024)