Gen-Searcher-SFT-10k; Gen-Searcher-RL-6k
收藏资源简介:
Gen-Searcher-SFT-10k和Gen-Searcher-RL-6k是由香港中文大学MMLab等机构构建的高质量搜索增强图像生成训练数据集,包含约16,000条样本。数据集通过精心设计的提示工程和深度网络搜索策略生成,涵盖20多个多样化领域,如动漫、艺术、物理等。数据创建过程涉及多轮代理搜索轨迹生成和基于Nano Banana Pro的图像合成,并经过Seed1.8模型评分和人工验证以确保质量。该数据集旨在解决知识密集型场景下的图像生成问题,为训练搜索增强型图像生成代理提供基础支持。
Gen-Searcher-SFT-10k and Gen-Searcher-RL-6k are high-quality search-augmented image generation training datasets constructed by MMLab of The Chinese University of Hong Kong and other institutions, containing approximately 16,000 samples in total. These datasets are generated via meticulously designed prompt engineering and deep web search strategies, covering more than 20 diverse domains such as animation, art, physics and other fields. The data creation process involves multi-round agent search trajectory generation and image synthesis based on Nano Banana Pro, and has been scored by the Seed1.8 model and manually verified to ensure quality. These datasets aim to address the image generation problem in knowledge-intensive scenarios, providing fundamental support for training search-augmented image generation AI agents.




