Instructions to use BruceYg/ATPO-G-Qwen2.5-VL-7B-r5 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use BruceYg/ATPO-G-Qwen2.5-VL-7B-r5 with Transformers:
# pip install -U transformers accelerate # Load model directly from transformers import AutoProcessor, AutoModelForMultimodalLM processor = AutoProcessor.from_pretrained("BruceYg/ATPO-G-Qwen2.5-VL-7B-r5") model = AutoModelForMultimodalLM.from_pretrained("BruceYg/ATPO-G-Qwen2.5-VL-7B-r5", device_map="auto") - Notebooks
- Google Colab
- Kaggle
ATPO-G-Qwen2.5-VL-7B-r5
ATPO-G-Qwen2.5-VL-7B-r5 is based on Qwen2.5-VL-7B-Instruct and fine-tuned for multi-label video safety detection using Adaptive Tversky Policy Optimization (ATPO). The ATPO-G variant uses Global Adaptive Tversky Reward (G-ATR), which adapts false-positive and false-negative penalties during reinforcement learning to steer the precision-recall trade-off.
The r5 suffix indicates a global training target FN/FP ratio of 5.0, favoring higher overall precision to reduce false positives, with a trade-off in recall. G-ATR applies this target to errors aggregated across categories, as described in the paper, Section 4.4.1.
This repository contains merged BF16 model weights, tokenizer files, and processor configurations.
- Paper: Controllable Multi-label Video Safety Detection via Adaptive Tversky Policy Optimization
- Code: BruceYg/ATPO
- Project page: ATPO
Citation
@inproceedings{yang2026controllable,
title={Controllable Multi-label Video Safety Detection via Adaptive Tversky Policy Optimization},
author={Guangyu Yang and Jingbiao Mei and Mingsheng Sun and Jinghong Chen and Yingtong Bu and Pengda Qin and Da Chen and Bill Byrne},
booktitle={The Fortieth Annual Conference on Neural Information Processing Systems},
year={2026},
url={https://arxiv.org/abs/2610.02019}
}
- Downloads last month
- 19