Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

MODE: Modality-Decomposed Expert-Level Mixed-Precision Quantization for MoE Multimodal LLMs

About

Mixture-of-Experts Multimodal Large Language Models (MoE-MLLMs) offer remarkable performance but incur prohibitive GPU memory costs, making compression essential. Among PTQ methods, expert-level mixed-precision quantization has proven effective for MoE-LLMs, yet suffers notable degradation on MoE-MLLMs due to two overlooked biases in expert importance estimation. (1) At the cross-modal level, the numerical dominance of vision tokens causes expert selection frequency to be dominated by vision tokens, masking experts that are critical to the text modality; (2) at the intra-vision level, the large proportion of redundant vision tokens further skew frequency statistics, obscuring experts critical for informative visual content. To bridge gaps, we propose MODE, a modality-decomposed expert-level mixed-precision quantization framework for MoE-MLLMs that decomposes expert selection frequency by modality, filters redundant vision tokens to obtain denoised visual frequency, and further evaluates quantization sensitivity per modality as a complementary signal to frequency-based estimation. These signals are integrated into an Integer Linear Programming formulation to assign per-expert bit-widths under a given budget. Extensive experiments show that MODE is particularly well-suited for MoE-MLLMs, limiting average performance loss to within 2.9% at W3A16, with larger gains at the extreme 2-bit setting.

Yuanteng Chen, Peisong Wang, Zhilei Liu, Nanxin Zeng, Yuantian Shao, Shiqiang Lang, Tao Liu, Chuangyi Li, Qinghao Hu, Gang Li, Jing Liu, Jian Cheng• 2026

Related benchmarks

TaskDatasetResultRank
Multimodal UnderstandingMMStar
Accuracy55.68
511
Object Hallucination EvaluationPOPE
Accuracy89.44
259
Visual Question AnsweringVizWiz
Accuracy69.42
193
Chart Question AnsweringChartQA
Accuracy87.26
165
Information Visual Question AnsweringInfoVQA
Accuracy79.87
159
Text-based Visual Question AnsweringTextVQA
TextVQA Accuracy86.24
141
Multimodal UnderstandingMMBench
Accuracy82.56
137
Visual Question AnsweringGQA
GQA Score60.15
75
Multimodal UnderstandingMMMU
Accuracy (MMMU)57.33
73
Multimodal EvaluationMME-R
Score50.77
30
Showing 10 of 17 rows

Other info

Follow for update