MODF-SIR: A Multi-agent Omni-modal Distilled Framework for Social Intelligence Reasoning

The article presents MODF-SIR, a multi-agent collaborative framework leveraging a lightweight Multimodal Large Language Model (MLLM) for social intelligence reasoning, enhanced through knowledge distillation. Key features include the integration of Test-Time Adaptation (TTA) and Low-Rank Adaptation (LoRA) for instance-level reasoning, which allows for the effective extraction and representation of long-tail events while preventing noise interference. The framework achieves state-of-the-art results across various benchmarks, making it a significant advancement for practitioners focusing on social intelligence in AI applications.