LLM quantization

MixLLM

MixLLM: LLM Quantization with Global Mixed-precision between Output-features and Highly-efficient System Design

Superseded baseline#79 of 80 most-superseded · first seen Dec 19, 2024

Cited as a baseline — critiqued by newer work, not yet beaten on a benchmark here

1 papers critique it · 0 beat it on benchmarks

What papers say

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

its reliance on naive thresholding and neglect of activation distortion in weight-activation joint quantization often lead to suboptimal performance
SignRoundV2: Closing the Performance Gap in Extremely Low-Bit Post-Training Quantization for LLMs

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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.