近日，中国人工智能研究机构——智源研究院在其官方微博上宣布，已经成功开源了其最新研发的新一代多模态基础模型——Emu2。
据悉，Emu2是通过大规模自回归生成式多模态预训练而得到的成果，它的出现标志着在多模态上下文学习方面取得了重大突破。与其他主流的多模态预训练大模型相比，如Flamingo-80B和IDEFICS-80B，Emu2在少样本多模态理解任务上的表现更为出色。
在VQAv2、OKVQA、MSVD、MM-Vet和TouchStone等多项少样本理解、视觉问答、主体驱动图像生成的任务中，Emu2均实现了最优性能，这无疑进一步证明了其在多模态处理领域的领先地位。
此次智源研究院的开源行动，不仅为科研人员提供了最新的研究成果和技术支持，也推动了多模态技术的发展和应用，对于提升我国在全球人工智能领域的竞争力具有重要意义。
英语如下：
News Title: “Zhizyuan Research Institute releases Emu2: The Next Generation Multimodal Model Leads to New Breakthroughs in Few-shot Understanding”
Keywords: Zhizyuan Research Institute, open source release, Emu2 model
News Content: Title: Zhizyuan Research Institute Successfully Open Sources Its Latest Multimodal Foundation Model – Emu2
Recently, the Chinese artificial intelligence research institution – Zhizyuan Research Institute announced on its official Weibo account that it has successfully open sourced its latest research and development of the next generation multimodal foundation model – Emu2.
It is understood that Emu2 is the result of large-scale autoregressive generative multimodal pre-training. Its emergence marks a major breakthrough in multi-modal context learning. Compared with other mainstream multimodal pre-trained large models such as Flamingo-80B and IDEFICS-80B, Emu2 performs better in few-shot multi-modal understanding tasks.
In various tasks such as VQAv2, OKVQA, MSVD, MM-Vet, and TouchStone involving few-shot understanding, visual question answering, and subject-driven image generation, Emu2 has achieved optimal performance, which further confirms its leading position in the field of multimodal processing.
This open-source action by Zhizyuan Research Institute not only provides the latest research results and technical support for researchers but also drives the development and application of multimodal technology, which is of great significance for enhancing China’s competitiveness in the global artificial intelligence field.
【来源】https://mp.weixin.qq.com/s/Xf4xBzYwubVd8Lpw68ikDA
Views: 1