86 lines
5.5 KiB
Markdown
86 lines
5.5 KiB
Markdown
---
|
|
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
|
|
@article{yang2026paravt,
|
|
title={ParaVT: Taming the Tool Prior Paradox for Parallel Tool Use in Agentic Video Reinforcement Learning},
|
|
author={Yang, Zuhao and Zhang, Kaichen and Wang, Sudong and Wu, Keming and Yang, Zhongyu and Li, Bo and Qi, Xiaojuan and Lu, Shijian and Li, Xingxuan and Bing, Lidong},
|
|
journal={arXiv preprint arXiv:2605.20342},
|
|
year={2026}
|
|
}
|
|
```
|
|
|
|
## 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.
|