File size: 5,645 Bytes
e051198 68e7dc7 e051198 38409d2 e399087 b3e2ae8 e051198 e399087 e051198 68e7dc7 e051198 b3e2ae8 e051198 b3e2ae8 e051198 b3e2ae8 e051198 58a9fe1 07f786f e051198 38409d2 e051198 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 | ---
base_model:
- Qwen/Qwen3-VL-8B-Instruct
datasets:
- ParaVT/ParaVT-Parquet
- ParaVT/ParaVT-Source
license: apache-2.0
library_name: transformers
pipeline_tag: video-text-to-text
language:
- en
tags:
- video
- long-video
- reasoning
- tool-calling
- agentic-rl
- grpo
- multimodal
---
# ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning
<div align="center">
[](https://arxiv.org/abs/2605.20342)
[](https://evolvinglmms-lab.github.io/ParaVT/)
[](https://github.com/EvolvingLMMs-Lab/ParaVT)
[](https://huggingface.co/datasets/ParaVT/ParaVT-Parquet)
[](https://huggingface.co/datasets/ParaVT/ParaVT-Source)
[](https://huggingface.co/papers/2605.20342)
</div>
## Overview
Training large multimodal models (LMMs) via reinforcement learning to natively invoke video-processing tools (such as temporal cropping) has become a promising route to long-video understanding. Existing native-RL methods, however, dispatch tool calls sequentially (one per turn): a single wrong crop propagates errors without peer correction, multi-turn calls corrupt context, and inference cost scales linearly with the number of turns.
**ParaVT** is the first multi-agent end-to-end RL-trained framework for **Para**llel **V**ideo **T**ool calling: it dispatches multiple time-window crops in a single turn for cleaner context and better fault tolerance. Applying standard RL to ParaVT surfaces an obstacle we term the *Tool Prior Paradox*, where the pretrained tool priors that enable tool exploration also destabilize cold-started structural format and expose a skip-tool reward shortcut under temperature sampling. We address this with **PARA-GRPO** (Parseability-Anchored and Ratio-gAted GRPO): a targeted format reward applied only at the structural-token positions most prone to collapse, and a per-prompt frame-budget randomization that creates training prompts where calling the tool yields a measurable reward signal over skipping it.
## Model Card
This repository hosts the final post-RL checkpoint (`ParaVT-8B`), obtained by running PARA-GRPO on top of the cold-start SFT checkpoint [`mwxely/ParaVT-8B-SFT`](https://huggingface.co/mwxely/ParaVT-8B-SFT). The base architecture is `Qwen3VLForConditionalGeneration`, identical to [`Qwen/Qwen3-VL-8B-Instruct`](https://huggingface.co/Qwen/Qwen3-VL-8B-Instruct); only the language-model weights are updated.
| Field | Value |
|---|---|
| Architecture | `Qwen3VLForConditionalGeneration` |
| Parameters | 8 B |
| Base model | `Qwen/Qwen3-VL-8B-Instruct` |
| Training stages | Cold-start SFT (500 steps) → PARA-GRPO RL (54 steps) |
| Training data | [`ParaVT/ParaVT-Parquet`](https://huggingface.co/datasets/ParaVT/ParaVT-Parquet) (`sft` + `rl` configs) |
| Source videos | [`ParaVT/ParaVT-Source`](https://huggingface.co/datasets/ParaVT/ParaVT-Source) |
| Native tool | Temporal cropping (start time, end time, optional sub-frame count) |
## Usage
`ParaVT-8B` is a drop-in `transformers` / `vllm` model for video-text-to-text. The full evaluation driver, prompt templates, and reproduction scripts live in the [ParaVT GitHub repository](https://github.com/EvolvingLMMs-Lab/ParaVT); please refer to it for the exact environment that produced the reported numbers.
```bash
# Reproduce the headline numbers (after installing the eval venv)
git clone https://github.com/EvolvingLMMs-Lab/ParaVT.git && cd ParaVT
cp .secrets.env.example .secrets.env && $EDITOR .secrets.env
bash scripts/setup_env.sh eval
PARAVT_EVAL_MODEL=ParaVT/ParaVT-8B \
bash paravt/eval/scripts/reproduce_paravt_8b.sh
```
For inference outside the eval driver, treat the model exactly like `Qwen/Qwen3-VL-8B-Instruct`: vLLM `--model ParaVT/ParaVT-8B`, the same tokenizer, the same chat template. The agentic system prompt and the tool schema used during PARA-GRPO are documented in [`paravt/eval/configs/withtool.yaml`](https://github.com/EvolvingLMMs-Lab/ParaVT/blob/main/paravt/eval/configs/withtool.yaml) and [`paravt/eval/utils.py`](https://github.com/EvolvingLMMs-Lab/ParaVT/blob/main/paravt/eval/utils.py).
## Citation
If you find ParaVT useful for your research and applications, please cite:
```bibtex
@misc{yang2026paravt,
title={{ParaVT}: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning},
author={Zuhao Yang and Kaichen Zhang and Sudong Wang and Keming Wu and Zhongyu Yang and Bo Li and Xiaojuan Qi and Shijian Lu and Xingxuan Li and Lidong Bing},
year={2026},
eprint={2605.20342},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```
## Acknowledgements
ParaVT builds on the [LongVT](https://github.com/EvolvingLMMs-Lab/LongVT) (CVPR 2026) framework for native video tool calling, the [`lmms-engine`](https://github.com/EvolvingLMMs-Lab/lmms-engine) cold-start SFT infrastructure, the [`AReaL`](https://github.com/inclusionAI/AReaL) RL training stack, and the [`lmms-eval`](https://github.com/EvolvingLMMs-Lab/lmms-eval) evaluation harness. We thank the maintainers of all of the above.
|