GenSearcher/Train-Data
收藏资源简介:
Gen-Searcher训练数据包含两个高质量数据集:Gen-Searcher-SFT-10k和Gen-Searcher-RL-6k,这些数据集旨在支持需要复杂现实世界知识的图像生成任务。数据集包含多样化的搜索密集型提示和相应的真实合成图像,用于不同的训练阶段。具体包括用于监督微调(SFT)的`sft_data.json`、用于强化学习(RL)的`rl_data.json`以及包含所有16k文本-图像对的`all_data.json`。这些数据支持Gen-Searcher项目,该项目是一个多模态深度研究代理,能够进行网络搜索、浏览证据、推理多源信息并在生成前搜索视觉参考,从而实现更准确和最新的图像合成。
Gen-Searcher training data comprises two high-quality datasets: Gen-Searcher-SFT-10k and Gen-Searcher-RL-6k. These datasets are designed to support image generation tasks that require complex real-world knowledge. The datasets contain diverse search-intensive prompts and their corresponding authentic synthetic images for different training stages. Specifically, they include `sft_data.json` for supervised fine-tuning (SFT), `rl_data.json` for reinforcement learning (RL), and `all_data.json` which encompasses all 16k text-image pairs. These datasets support the Gen-Searcher project, a multimodal deep research agent capable of performing web searches, browsing evidence, reasoning across multi-source information, and searching for visual references prior to generation, thereby enabling more accurate and up-to-date image synthesis.




