Search papers, labs, and topics across Lattice.
ParaVT is introduced as a novel multi-agent RL framework that enables parallel video tool calling, addressing limitations of sequential tool use in long-video understanding by dispatching multiple time-window crops in a single turn. The paper identifies and analyzes the "Tool Prior Paradox," where strong pretrained tool priors in LMMs lead to both effective tool exploration and structural format instability during RL fine-tuning. To mitigate this, they propose PARA-GRPO, which incorporates targeted format rewards and frame-budget randomization, achieving a 7.9% average improvement over the Qwen3-VL baseline across six long-video understanding benchmarks and significantly improving format compliance.
RL fine-tuning LMMs for tool use can collapse structural formats due to strong pretrained tool priors, but a surprisingly simple fix of targeted format rewards and frame-budget randomization can restore stability and boost performance.
Training large multimodal models (LMMs) via reinforcement learning (RL) to natively invoke video-processing tools (e.g., cropping) has become a promising route to long-video understanding. However, existing native-RL methods dispatch tool calls sequentially (i.e., one per turn): a single wrong crop propagates errors without peer correction, multi-turn tool calls corrupt context, and inference cost scales linearly with the number of turns. We introduce ParaVT, the first multi-agent end-to-end RL-trained framework for Parallel Video Tool calling, dispatching multiple time-window crops in a single turn for cleaner context and better fault tolerance. Yet applying standard RL to ParaVT reveals an obstacle we term the Tool Prior Paradox: the pretrained tool priors that enable tool exploration also destabilize cold-started structural format and expose the skip-tool reward shortcut under temperature sampling. A cross-model contrast on a weaker-prior LMM supports this claim: format stays stable but RL elicits zero tool calls, indicating that prior strength is the shared driver of both format collapse and tool exploration. We propose PARA-GRPO (Parseability-Anchored and Ratio-gAted GRPO), which augments standard RL with two complementary mechanisms: (i) a targeted format reward applied only at the structural-token positions most prone to collapse, and (ii) a per-prompt frame-budget randomization that creates training prompts where calling the tool yields a measurable reward signal over skipping it. Across six long-video understanding benchmarks, ParaVT improves over the Qwen3-VL baseline by +7.9% on average, with PARA-GRPO lifting training-time format compliance from 0.13 to 0.64. As tool capabilities become increasingly internalized in modern LMMs, RL must cooperate with the resulting priors, and ParaVT offers a general recipe for agentic RL. Code, data, and model weights are publicly available.
