LUT Quantization
A non-uniform quantization implementation in which codebooks are pre-stored for weight or activation blocks and quantization matches the nearest centroid by table lookup. It can flexibly approximate arbitrary distributions and suits scenarios with uneven weights or outliers, but requires extra storage and custom inference kernels.
Timeline (last 90 days)
CodeQuant2 is a clustering-based lookup table (LUT) quantization framework designed specifically for low-precision large reasoning models.
Lookup table (LUT) quantization is a non-uniform quantization implementation that pre-stores a codebook, looks up the table to find the nearest center index during quantization, and retrieves the center value during dequantization.
All Facts (2)
CodeQuant2 is a clustering-based lookup table (LUT) quantization framework designed specifically for low-precision large reasoning models.
50%UnverifiedLookup table (LUT) quantization is a non-uniform quantization implementation that pre-stores a codebook, looks up the table to find the nearest center index during quantization, and retrieves the center value during dequantization.
50%