当前，像ChatGPT之类的LLM可谓是“风靡全球”。有读者留意到，几乎所有LLM都还是用最初的Multi-Head Scaled-Dot Attention，近年来大量的Efficient工作...| spaces.ac.cn
在去年的文章《Transformer升级之路：2、博采众长的旋转式位置编码》中，笔者提出了旋转位置编码（RoPE），当时的出发点只是觉得用绝对位置来实现相对位置是一件“很好玩的事情”，并没料到其...| spaces.ac.cn
在之前的文章《Transformer升级之路：2、博采众长的旋转式位置编码》中我们提出了旋转式位置编码RoPE以及对应的Transformer模型RoFormer。由于笔者主要研究的领域还是NL...| spaces.ac.cn
上一篇文章中，我们对原始的Sinusoidal位置编码做了较为详细的推导和理解，总的感觉是Sinusoidal位置编码是一种“想要成为相对位置编码的绝对位置编码”。一般来说，绝对位置编码具有实现...| spaces.ac.cn
2017年中，有两篇类似同时也是笔者非常欣赏的论文，分别是FaceBook的《Convolutional Sequence to Sequence Learning》和Google的《Atten...| spaces.ac.cn
RoPER, in addition to using relative positions in the attention score calculation, adds relative positional information explicitly to value embeddings.| labml.ai research blog
Our ability to use deep learning approaches to decipher neural activity would likely benefit from greater scale, in terms of both model size and datasets. However, the integration of many neural recordings into one unified model is challenging, as each recording contains the activity of different neurons from different individual animals. In this paper, we introduce a training framework and architecture designed to model the population dynamics of neural activity across diverse, large-scale n...| papers.cool
This is an implementation of RoPER which adds relative distance information to embeddings on top of RoPE introduced in RoFormer: Enhanced Transformer with Rotary Position Embedding| Rotary Positional Embeddings with Relative distance (RoPER)
In this work, we propose Retentive Network (RetNet) as a foundation architecture for large language models, simultaneously achieving training parallelism, low-cost inference, and good performance. We theoretically derive the connection between recurrence and attention. Then we propose the retention mechanism for sequence modeling, which supports three computation paradigms, i.e., parallel, recurrent, and chunkwise recurrent. Specifically, the parallel representation allows for training parall...| papers.cool
前几天，幻方发布的DeepSeek-V2引起了大家的热烈讨论。首先，最让人哗然的是1块钱100万token的价格，普遍比现有的各种竞品API便宜了两个数量级，以至于有人调侃“这个价格哪怕它输出乱...| spaces.ac.cn
持续将“Transformer升级之路”系列关注到本篇的读者，想必都已经对旋转位置编码（RoPE）有所了解。简单来说，RoPE是施加在Attention的Query（$\boldsymbol{Q...| spaces.ac.cn
事实上，除了写博客内容，在这几年里，笔者是花了相当一部分时间来做科学空间的“表面功夫”，为此还专门学了一点php、css和js。虽然不敢说精益求精，但总体来说网站的浏览体验应该比前几年要好得多。...| spaces.ac.cn