[
    {
        "id": "osp-22445",
        "type": "article-journal",
        "title": "Text-Centric Post-Training for Omni-Modal Reasoning",
        "author": [
            {
                "family": "Cheng",
                "given": "Ziyang"
            },
            {
                "family": "Wang",
                "given": "Yuhao"
            },
            {
                "family": "Liu",
                "given": "Hongcheng"
            },
            {
                "family": "Wu",
                "given": "Qimin"
            },
            {
                "family": "Fan",
                "given": "Jingru"
            },
            {
                "family": "Qian",
                "given": "Chen"
            },
            {
                "family": "Wang",
                "given": "Yanfeng"
            },
            {
                "family": "Wang",
                "given": "Yu"
            }
        ],
        "URL": "https://omanscience.com/en/articles/text-centric-post-training-for-omni-modal-reasoning",
        "language": "en",
        "issued": {
            "date-parts": [
                [
                    2026
                ]
            ]
        },
        "abstract": "Improving joint audio-visual reasoning in Omni Large Language Models typically incurs substantial data construction and training costs. Our diagnostics reveal multi-hop reasoning difficulties despite correct answers to all corresponding single-hop questions and suggest partial decoupling in the local optimization of perception and reasoning objectives. This motivates post-training with different emphases on these capabilities. Text-only reasoning training yields gains across data sources, model scales, and families. With the best-performing text-only configuration, supervised fine-tuning followed by reinforcement learning (RL) raises Qwen2.5-Omni-7B's geometric mean of nine reasoning scores by 25.83% over the base model, outperforming the complete native audio-visual route with 56.6% fewer GPU-hours. Training on data synthesized entirely by a text-only LLM raises this geometric mean by 21.01% without audio-visual data in construction or training. However, text-only training degrades perception. We therefore propose a text-centric post-training paradigm: text-only training provides the main reasoning optimization, and reduced-data native audio-visual RL then refines perception. Refinement uses about 90% fewer input tokens than full-data audio-visual RL, restores perception above the base level, and retains 93.5% of the best-performing text-only pipeline's reasoning gain."
    }
]