一个直观的解释是训练数据不足，但更本质的问题在于表示空间不匹配。已有研究表明，LLM 已经在统一的语义空间中编码了丰富的跨语言知识，并且在处理多语言文本时会专门「经过」这个统一语义空间（如英语表示空间）。这意味着， LLM ...

打破多模态视觉+语言拼接套路！ 腾讯开源Penguin-VL，直接用纯文本LLM训视觉编码器。 这项研究跳出了先有传统视觉 backbone，再接语言模型的常规路径，直接从text-only LLM初始化vision encoder。 并在2B/8B紧凑参数规模下的文档理解、长视频时序定位等复杂任务中表现出 ...

SK Telecom has unveiled a universal document interpretation technology for vision-language model (VLM) and large language model (LLM) training, based on its proprietary large language model, A.Dot X ...

Moonshot AI today released Kimi-K2.6, the latest addition to its popular Kimi series of open-source large language models.

Transformer-based models have rapidly spread from text to speech, vision, and other modalities. This has created challenges for the development of Neural Processing Units (NPUs). NPUs must now ...

Artificial intelligence is becoming an increasingly significant asset for companies worldwide, especially as they integrate generative AI features like chatbots into their services. However, deploying ...