AntResearchNLP/ViLaSR-data
收藏Hugging Face2025-06-24 更新2025-10-18 收录
下载链接:
https://hf-mirror.com/datasets/AntResearchNLP/ViLaSR-data
下载链接
链接失效反馈官方服务:
资源简介:
ViLaSR-data数据集是用于强化视觉语言模型中的空间推理的,通过交织思考和视觉绘图的方法。数据集包含三个主要部分:VILASR-ColdStart-33k(冷启动提示生成的初始数据),VILASR-RRS-8k(使用反思性拒绝采样 refined 的数据),VILASR-RL-40k(通过强化学习增强的数据)。
The ViLaSR-data dataset is designed for reinforcing spatial reasoning in vision-language models through interwoven thinking and visual drawing. The dataset consists of three main components: VILASR-ColdStart-33k (initial data generated from cold-start prompts), VILASR-RRS-8k (data refined using reflective rejection sampling), and VILASR-RL-40k (data enhanced with reinforcement learning).
提供机构:
AntResearchNLP



