Is EfficientQAT superseded?

EfficientQAT (LLM quantization): superseded — cited as a baseline and beaten by newer methods. 1 paper(s) critique it, 2 beat it on benchmarks — #22 of 80 most-superseded. Sub-problem: cluster led by RTN. Newer alternatives in the same sub-problem include STaR-Quant, Timestep-Aware SVDQuant-GPTQ, BWLA, Bit-by-Bit, Benford-Quant.

Method Drift›LLM quantization

Superseded baseline#22 of 80 most-superseded

EfficientQAT

EfficientQAT: Efficient Quantization-Aware Training for Large Language Models

LLM quantization · first seen Jul 10, 2024

superseded — cited as a baseline and beaten by newer methods

1 papers critique it · 2 beat it on benchmarks

What papers say

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

“The best quantization-aware training method, EfficientQAT, still suffers a 9.1-point decline in average accuracy. Our method dramatically narrows the 2-bit quantization gap to full precision to just 3.4 points, outperforming the best QAT method by 5.7 points and the vector quantization method by 7.1 points.”
— ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization

Beaten on benchmarks

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

Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-2 7B, w2a16]
6.50 vs 7.39
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-3.2-3B, w2a16]
11.02 vs 13.31
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-2 7B, w2a2]
7.72 vs 9.71
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-3.2-3B, w2a2]
13.87 vs 20.19
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-3.2-1B, w2a16]
16.13 vs 21.48
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
Bit-by-Bit (Ours) beats EfficientQAT · WikiText2 PPL [LLaMA-3-8B, w2a16]
8.32 vs 11.17
Bit-by-Bit: Progressive QAT Strategy with Outlier Channel Splitting for Stable Low-Bit LLMs
ApiQ beats EfficientQAT · WikiText2 PPL [LLaMA-2-7B, 3-bit]
5.77 vs 5.81
Enhancing Ultra-Low-Bit Quantization of Large Language Models Through Saliency-Aware Partial Retraining

Newer alternatives

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