{"id":15798,"date":"2024-07-18T08:42:57","date_gmt":"2024-07-18T00:42:57","guid":{"rendered":"https:\/\/www.1ai.net\/?p=15798"},"modified":"2024-07-18T08:42:57","modified_gmt":"2024-07-18T00:42:57","slug":"%e6%99%ba%e6%ba%90%e7%a0%94%e7%a9%b6%e9%99%a2%e6%8e%a8%e5%87%ba%e6%96%b0%e4%b8%80%e4%bb%a3%e6%97%a0%e7%bc%96%e7%a0%81%e5%99%a8%e8%a7%86%e8%a7%89%e8%af%ad%e8%a8%80%e5%a4%9a%e6%a8%a1%e6%80%81%e5%a4%a7","status":"publish","type":"post","link":"https:\/\/www.1ai.net\/en\/15798.html","title":{"rendered":"AI Research Institute launches a new generation of encoder-free visual language multimodal large model EVE"},"content":{"rendered":"

Recently,Multimodal large model<\/a>The research and application of multimodal models have made significant progress. Foreign companies such as OpenAI, Google, and Microsoft have launched a series of advanced models, and domestic institutions such as Zhipu AI and Jieyuexingchen have also made breakthroughs in this field. These models usually rely on visual encoders to extract visual features and combine them with large language models, but there is a problem of visual induction bias caused by training separation, which limits the deployment efficiency and performance of large multimodal models.<\/p>\n

To solve these problems,AI Research Institute<\/a>In collaboration with Dalian University of Technology, Peking University and other universities, we launched a new generation of encoder-free visual language model EVE. EVE integrates visual-language representation, alignment and reasoning into a unified pure decoder architecture through refined training strategies and additional visual supervision. Using public data, EVE performs well in multiple visual-language benchmarks, approaching or even outperforming mainstream encoder-based multimodal methods.<\/p>\n

\"\"<\/p>\n

The main features of EVE include:<\/p>\n