智源研究院日前宣布开源发布新一代多模态基础模型Emu2，该模型通过大规模自回归生成式多模态预训练，显著推动多模态上下文学习能力的突破。Emu2在少样本多模态理解任务上大幅超越Flamingo-80B、IDEFICS-80B等主流多模态预训练大模型，在包括VQAv2、OKVQA、MSVD、MM-Vet、TouchStone在内的多项少样本理解、视觉问答、主体驱动图像生成等任务上取得最优性能。
Emu2的发布引起了业内的高度关注，它标志着智源研究院在多模态基础模型领域取得了重要突破，进一步提升了我国在人工智能领域的影响力。
新闻翻译：
Title:智源研究院发布新一代多模态基础模型 Emu2， significantly outperforms mainstream models
Keywords:智源研究院，多模态基础模型，Emu2，breakthrough
News content:
Zhiyuan Research Institute recently announced the open source release of a new generation of multi-modal basic model Emu2. This model, through large-scale self-reliable generation and multi-modal pre-training, significantly promotes the breakthrough of multi-modal context learning ability. Emu2 significantly outperforms Flamingo-80B, IDEFICS-80B, and other mainstream multi-modal pre-trained large models on the task of few-sample multi-modal understanding, and achieves the optimal performance on multiple tasks such as visual question answering, object-oriented image generation, and others.
Emu2’s release has attracted widespread attention in the field, and it marks a significant breakthrough in the field of multi-modal basic models, further enhancing China’s influence in the field of artificial intelligence.
【来源】https://mp.weixin.qq.com/s/Xf4xBzYwubVd8Lpw68ikDA
Views: 1