Scholar Feed

LLM quantization

MBQ

MBQ: Modality-Balanced Quantization for Large Vision-Language Models

Superseded baseline#18 of 80 most-superseded · first seen Dec 27, 2024

Superseded — cited as a baseline and beaten by newer methods

2 papers critique it · 2 beat it on benchmarks

What papers say

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

Existing approaches~li2025mbq typically learn a single transformation shared by both modalities across all channels, where the cross-modal heterogeneity can severely distort the optimization objective.
— Breaking Modality Heterogeneity in Low-Bit Quantization for Large Vision-Language Models
“these methods either require expensive parameter fine-tuning xie2024qslaw, specialized manipulation at inference time yu2025mquant, or rely on a suboptimal grid search li2025mbq, failing to offer an efficient and effective solution for both calibration and inference”
— VLMQ: Efficient Post-Training Quantization for Large Vision-Language Models via Hessian Augmentation

Beaten on benchmarks

Head-to-head results where a newer method reports beating MBQ. 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Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.