VideoJev 对视频答案直接评分，报告称问答速度提高100倍
由 Nanjing University 牵头的项目发布了 4B 和 35B-A3B 模型；其速度测试测量的是在 H200 GPUs 上的模型前向时间，而不是端到端响应时间。
By Ryan Merket · Published
Primary source: X
为何重要
VideoJev 通过对选项打分并重用视频计算而不是每次生成文本，来测试视频代理是否能回答重复的、结构化的问题。它的速度声明仅限于模型前向推理的基准计时，未对端到端部署成本进行测试。
Researchers led by Nanjing University have released VideoJev, a pair of video-understanding models that score candidate answers and event times directly instead of generating a written response. In an October 14th post on X, Aran Komatsuzaki (@arankomatsuzaki) highlighted the project's claim of more than 100x faster video question answering and released 4B and 35B-A3B variants.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.