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

Stepwise Token Selection for Efficient Multimodal Large Language Models

About

In multimodal large language models (MLLMs), inference cost is largely dominated by the visual token prefix rather than the language backbone, making token reduction a key factor for improving efficiency. Existing approaches typically assign independent importance scores to visual tokens and retain a fixed number of top-ranked tokens, implicitly assuming token independence and a uniform compression ratio across inputs. In this work, we reformulate visual token pruning as a sequential decision-making process. Specifically, we introduce a pointer-style selection mechanism that iteratively chooses informative tokens, conditioning each decision on previously selected ones, and dynamically determines when to stop via a learned termination action. This enables joint optimization of both the selected subset and its size. To enable end-to-end training under standard language modeling objectives, we design a differentiable relaxation based on a variance-preserving noise interpolation scheme, allowing gradients to propagate through the discrete selection process. Extensive experiments on LLaVA-v1.5-7B and Qwen2.5-VL-7B demonstrate that our approach consistently outperforms fixed-ratio baselines across different compression levels. Under aggressive pruning that removes 88.9% of visual tokens, our method preserves 94.6% of the original accuracy while achieving a 1.88x speed-up in prefill latency.

Landi He, Shawn Young, Lijian Xu• 2026

Related benchmarks

TaskDatasetResultRank
Object Hallucination EvaluationPOPE
Accuracy85.3
259
Text-based Visual Question AnsweringVQAText
Accuracy77.2
89
Multimodal UnderstandingMMB
Accuracy80.5
63
Multimodal EvaluationLLaVA Evaluation Suite 7B v1.5 (test)
GQA59.5
59
Science Question AnsweringSQA
Accuracy (SQA)86.7
52
Multimodal PerceptionMME
Total Score2.30e+3
42
Multimodal Visual Question AnsweringGQA, MMB, POPE, and MME Average Retention LLaVA-v1.5-7B (test)
Accuracy94.6
4
Showing 7 of 7 rows

Other info

Follow for update