遇见数据集

alibaba-pai/OmmiThoughtV_Raw_1.8M

收藏
Hugging Face2026-01-26 更新2025-12-20 收录
官方服务:

资源简介:

OmniThoughtV是基于Finevision数据集和阿里云人工智能平台(PAI)的蒸馏工具包EasyDistill蒸馏得到的大规模多模态长思考数据集。该数据集构建了一套透明、可复现的数据蒸馏机制,实现了多模态推理思维链的高效构建。使用该数据集微调小模型可以成功赋予其更强的推理能力,并提升其在下游任务的表现。

OmniThoughtV is a large-scale multimodal long-thought dataset distilled from the Finevision dataset and the Alibaba Cloud AI Platform (PAI) distillation toolkit EasyDistill. The dataset establishes a transparent and reproducible data distillation mechanism, enabling efficient construction of multimodal reasoning chains. Fine-tuning small models with this dataset can successfully enhance their reasoning capabilities and improve their performance on downstream tasks.

提供机构:
alibaba-pai
二维码
社区交流群
二维码
科研交流群
商业服务