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

freePruner: A Training-free Approach for Large Multimodal Model Acceleration

About

Large Multimodal Models (LMMs) have demonstrated impressive capabilities in visual-language tasks but face significant deployment challenges due to their high computational demands. While recent token reduction methods show promise for accelerating LMMs, they typically require extensive retraining or fine-tuning, making them impractical for many state-of-the-art models, especially those with proprietary training data. We propose freePruner, a training-free token reduction approach that can be directly applied to any open-source LMM without additional training. Unlike existing methods that rely heavily on token merging operations, freePruner employs a two-stage token selection strategy: (1) identifying pivotal tokens that capture high-level semantic information using our designed contribution degree metric, and (2) selecting complementary tokens that preserve essential low-level visual details through attention pattern analysis. Extensive experiments demonstrate that freePruner achieves 2x acceleration while maintaining comparable performance across mainstream visual question-answering benchmarks in the training-free setting. Moreover, freePruner is orthogonal to and can be combined with other post-training acceleration techniques, such as post-training quantization, providing a practical solution for efficient LMM deployment.

Bingxin Xu, Yuzhang Shang, Yunhao Ge, Qian Lou, Yan Yan• 2024

Related benchmarks

TaskDatasetResultRank
Video Question AnsweringMSRVTT-QA
Accuracy59.5
513
Video Question AnsweringActivityNet-QA
Accuracy48.4
438
Video Question AnsweringMSVD-QA
Accuracy71.3
401
Object Hallucination EvaluationPOPE
Accuracy87.7
259
Visual Question AnsweringVQA v2
Accuracy77.6
257
Multimodal EvaluationMME
MME Score1.49e+3
179
Text-based Visual Question AnsweringTextVQA VQAT
Accuracy60
71
Multimodal EvaluationMM-Bench
Accuracy63.8
67
Visual Question AnsweringScienceQA SQA-I
Accuracy68.6
21
Showing 9 of 9 rows

Other info

Follow for update