Quantization method speeds up AI models on edge devices

Original: OrpQuant: Geometric Orthogonal Residual Projection for Multiplier-Free Power-of-Two Transformer Quantization

Source: arxiv.org ↗