Is SVDQuant superseded?

SVDQuant (LLM quantization): superseded — cited as a baseline and beaten by newer methods. 3 paper(s) critique it, 4 beat it on benchmarks — #10 of 80 most-superseded. Sub-problem: cluster led by SmoothQuant. Newer alternatives in the same sub-problem include OffQ, FAIR-Calib, InfoQuant, ConQuR, Activation Residual Hessian Quantization (ARHQ).

Method Drift›LLM quantization

Superseded baseline#10 of 80 most-superseded

SVDQuant

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

LLM quantization · 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superseded — cited as a baseline and beaten by newer methods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.

QuantVSR beats SVDQuant · PSNR [REDS4, W4A4]
23.31 vs 21.19
QuantVSR: Low-Bit Post-Training Quantization for Real-World Video Super-Resolution
QuantVSR beats SVDQuant · PSNR [SPMCS, W4A4]
22.76 vs 21.17
QuantVSR: Low-Bit Post-Training Quantization for Real-World Video Super-Resolution
LoRaQ beats SVDQuant · FID [PixArt-Σ (20 Steps), SINT4 4-4 precision]
16.9 vs 19.2
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · IR [PixArt-Σ (20 Steps), SINT4 4-4 precision]
0.898 vs 0.878
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · LPIPS [PixArt-Σ (20 Steps), SINT4 4-4 precision]
0.309 vs 0.323
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · FID [SANA-1.6B (20 Steps), SINT4 4-4 precision]
16.1 vs 19.3
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · IR [SANA-1.6B (20 Steps), SINT4 4-4 precision]
1.09 vs 0.935
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · LPIPS [SANA-1.6B (20 Steps), SINT4 4-4 precision]
0.182 vs 0.220
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · PSNR [SANA-1.6B (20 Steps), SINT4 4-4 precision]
19.2 vs 17.8
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · FID [MXINT4 4-4 precision]
15.4 vs 18.9
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · IR [MXINT4 4-4 precision]
0.901 vs 0.738
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization
LoRaQ beats SVDQuant · LPIPS [MXINT4 4-4 precision]
0.339 vs 0.424
LoRaQ: Optimized Low Rank Approximation for 4-bit Quantization

Newer alternatives

Recent methods in the same sub-problem, not yet superseded in the knowledge base.