喵~ 大家好呀，有个超棒的科技新消息要告诉你们哦！谷歌的DeepMind最近又放大招了，推出了两款超厉害的AI基础模型，叫做“Hawk”和“Griffin”呢，就像是新闻界的鹰眼和狮鹫，威风凛凛的说！
这次DeepMind的研究员们可真是脑洞大开，他们发明了一种叫做RG-LRU的新鲜玩意儿，这可是一种特别的门控线性循环层哦。这个RG-LRU层就像魔法一样，取代了以前多查询注意力（MQA）的方式，让信息处理变得更高效啦。
然后，他们就用这个新奇的循环块搭建了两个新模型。一个是“Hawk”，它巧妙地融合了MLP（多层感知机）和这个循环块，就像鹰一样敏锐地捕捉信息。另一个是“Griffin”，它不仅有MLP和循环块，还加上了局部注意力，就像狮鹫一样，既有广阔的视野，又能细致入微地观察。
这两个新模型的诞生，无疑为AI领域带来了新的风向标，让机器学习变得更加智能和灵活。是不是觉得科技世界每天都有新惊喜呢？喵，记得持续关注哦，精彩还在继续呢！
英语如下：
Title: “Google’s DeepMind Unveils Innovative Models Hawk and Griffin, Redefining the Era of AI Learning”
Keywords: DeepMind new models, Hawk&Griffin, RG-LRU layer
News Content: Meow~ Hi there, sweeties! I’ve got a fantastic tech update for you! Google’s DeepMind has just dropped some major innovation with two awesome AI foundation models called “Hawk” and “Griffin.” They’re like the superhero duo of the news world, don’t you think?
The DeepMind researchers really pushed the boundaries with this one. They created something called RG-LRU, a novel Gate-Controlled Linear Recurrent Unit. It’s like magic, replacing the old Multi-Query Attention (MQA) method and making information processing way more efficient.
With this nifty recurrent block, they built these new models. “Hawk” cleverly combines MLP (Multi-Layer Perceptron) with the block, just like an eagle, swiftly grabbing information. Then there’s “Griffin,” which not only has MLP and the block but also local attention, giving it a broad perspective like a griffin, yet with keen scrutiny.
These two models are game-changers in the AI field, making machine learning smarter and more flexible. Doesn’t it feel like technology surprises us every day? Meow, keep an eye out, because the excitement continues!
【来源】https://mp.weixin.qq.com/s/RtAZiEzjRWgqQw3yu3lvcg
Views: 1