Research map / Vision-language and multimodal / Multimodal LLMs
MMAT-1M: A Large Reasoning Dataset for Multimodal Agent Tuning
Tianhong Gao, Yannian Fu, Weiqun Wu, Haixiao Yue, Shanshan Liu, Gang Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| fine-tune with RR strategy on MMAT-1M | InternVL2.5-8B · 8 MLLM benchmarks (avg) | average score: 60.7 -> 63.4 | Improved |
| fine-tune with ORR strategy on MMAT-1M | InternVL2.5-8B · 8 MLLM benchmarks (avg) | average score: 60.7 -> 62.4 | Improved |
| fine-tune with RR on MMAT-1M | Llama-3.2-11B-Vision-Instruct · 8 MLLM benchmarks (avg) | average score: 52.2 -> 55.3 | Improved |
| fine-tune with RR on MMAT-1M | MiniCPM-V-2.6 · 8 MLLM benchmarks (avg) | average score: 58.0 -> 59.9 | Improved |
| fine-tune with RR on MMAT-1M | InternVL2.5-2B · Dyn-VQA (Query) | F1-Recall: 19.3 -> 30.9 | Improved |
| fine-tune with RR on MMAT-1M | Llama-3.2-11B-Vision-Instruct · Dyn-VQA (Query) | F1-Recall: 29.4 -> 38.0 | Improved |
| use RR instead of ORR | InternVL2.5-8B · OCRBench | score: 849 -> 839 | Hurt |
| use RR instead of ORR | InternVL2.5-2B · OCRBench | score: 819 -> 799 | Hurt |
| add API (tool) operators to RR (ablation) | Baseline-RR (InternVL2.5-8B) · avg of 9 benchmarks | average score: 57.3 -> 61.3 | Improved |
| add RAG retrieval to RR (ablation) | Baseline-RR (InternVL2.5-8B) · avg of 9 benchmarks | average score: 59.8 -> 61.3 | Improved |
| add SFT on MMAT-1M to RR (ablation) | Baseline-RR (InternVL2.5-8B) · avg of 9 benchmarks | average score: 55.0 -> 61.3 | Improved |
| add reflection to rationale (R to RR) (ablation) | Baseline-R (InternVL2.5-8B) · avg of 9 benchmarks | average score: 60.2 -> 61.3 | Improved |
Closest papers by meaning
- MC-Search: Evaluating and Enhancing Multimodal Agentic Search with Structured Long Reasoning Chains (ICLR 2026 · similarity 0.91)
- MME-CoT: Benchmarking Chain-of-Thought in Large Multimodal Models for Reasoning Quality, Robustness, and Efficiency (ICML 2025 · similarity 0.91)
- MDK12-Bench: A Multi-Discipline Benchmark for Evaluating Reasoning in Multimodal Large Language Models (AAAI 2026 · similarity 0.90)
- AvaTaR: Optimizing LLM Agents for Tool Usage via Contrastive Reasoning (NeurIPS 2024 · similarity 0.89)
- T1: A Tool-Oriented Conversational Dataset for Multi-Turn Agentic Planning (NeurIPS 2025 · similarity 0.89)
- Exploring the Limits of Sub-Billion Language Model Reasoners with Open Training Recipes (ICLR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
