现代搜索系统的核心挑战不仅在于从海量文档集合中检索相关信息，更在于对检索结果进行精准排序，确保用户能够快速、可靠且经济高效地获得所需信息。在面对不同重排序技术方案时，工程师们需要在延迟性能、硬件资源消耗、系统集成复杂度以及用户体验 ...

这项由马里兰大学和Meta公司联合完成的突破性研究发表于2025年5月28日的arXiv预印本平台（arXiv:2505.22664v1 [cs.CV]），论文题为《通过LLM替身实现零样本视觉编码器嫁接》(Zero-Shot Vision Encoder Grafting via LLM Surrogates)。该研究由Kaiyu Yue、Vasu Singla、Menglin ...

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 的多语言瓶颈不在 ...

LCLMs compress LLM context before decode — 8.8x faster at 16x compression, beating every KV cache method tested. Open-sourced by NYU and Columbia.

A vast majority of multi-modal AI systems function as a relay race. For example, an image will come in through the Vision ...

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 ...

For enterprise leaders aiming to decentralize their AI workloads, Gemma 4 12B offers a rare combination of edge-friendly ...

The company has announced the release of a new Gemma 4 model that fills a gap in the lineup that launched earlier this year.