ThinkOmni: Lifting Textual Reasoning to Omni-modal Scenarios via Guidance Decoding
This work provides a training-free and data-free solution for improving the complex reasoning abilities of omni-modal large language models, which is significant for researchers and developers facing high computational costs and data requirements in OLLM development.
This paper addresses the challenge of enhancing reasoning abilities in omni-modal large language models (OLLMs) without additional training. The proposed ThinkOmni framework, which is training-free and data-free, lifts textual reasoning to omni-modal scenarios by leveraging large reasoning models (LRMs) to guide OLLM decoding and adaptively balancing perception and reasoning signals. ThinkOmni achieves performance improvements on six multi-modal reasoning benchmarks, with main results of 70.2 on MathVista and 75.5 on MMAU.
Omni-modal reasoning is essential for intelligent systems to understand and draw inferences from diverse data sources. While existing omni-modal large language models (OLLM) excel at perceiving diverse modalities, they lack the complex reasoning abilities of recent large reasoning models (LRM). However, enhancing the reasoning ability of OLLMs through additional training presents significant challenges, including the need for high-quality data, task-specific adaptation, and substantial computational costs. To address these limitations, we propose ThinkOmni, a training-free and data-free framework that lifts textual reasoning to omni-modal scenarios. ThinkOmni introduces two key components: 1) LRM-as-a-Guide, which leverages off-the-shelf LRMs to guide the OLLM decoding process; 2) Stepwise Contrastive Scaling, which adaptively balances perception and reasoning signals without manual hyperparameter tuning. Experiments on six multi-modal reasoning benchmarks demonstrate that ThinkOmni consistently delivers performance improvements, with main results achieving 70.2 on MathVista and 75.5 on MMAU. Overall, ThinkOmni offers a flexible and generalizable solution for omni-modal reasoning and provides new insights into the generalization and application of reasoning capabilities.