Scholar Feed

LLM quantization

SVDQuant

SVDQuant: Absorbing Outliers by Low-Rank Components for 4-Bit Diffusion Models

Superseded baseline#10 of 80 most-superseded · first seen Nov 7, 2024

Superseded — cited as a baseline and beaten by newer methods

3 papers critique it · 4 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites SVDQuant as a baseline.

SVD-based approximations fail to preserve channel-wise outlier structures critical for contextual understanding.
— SpecQuant: Spectral Decomposition and Adaptive Truncation for Ultra-Low-Bit LLMs Quantization
“absorbing magnitude outliers into a high-precision branch to reduce the dynamic range and clipping errors of the quantized main branch”
— Technical Report: Activation Residual Hessian Quantization (ARHQ) for Low-Bit LLM Quantization
“However, SVDQuant does not explicitly account for the timestep-dependent activation distribution shift during diffusion denoising.”
— Timestep-Aware SVDQuant-GPTQ for W4A4 Quantization of Wan2.2-I2V

Beaten on benchmarks

Head-to-head results where a newer method reports beating SVDQuant. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.