An essential component of modern recurrent sequence models is the forget gate. While Transformers do not have an explicit recurrent form, we show that a forget gate can be naturally incorporated into Transformers by down-weighting the unnormalized attention scores in a data-dependent way. We name this attention mechanism the Forgetting Attention and the resulting model the Forgetting Transformer (FoX). We show that FoX outperforms the Transformer on long-context language modeling, length extr...| papers.cool
Scaling language models to handle longer input sequences typically necessitates large key-value (KV) caches, resulting in substantial memory overhead during inference. In this paper, we propose Tensor Product Attention (TPA), a novel attention mechanism that uses tensor decompositions to represent queries, keys, and values compactly, significantly shrinking KV cache size at inference time. By factorizing these representations into contextual low-rank components (contextual factorization) and ...| papers.cool
We propose novel attention architectures, Multi-matrix Factorization Attention (MFA) and MFA-Key-Reuse (MFA-KR). Existing variants for standard Multi-Head Attention (MHA), including SOTA methods like MLA, fail to maintain as strong performance under stringent Key-Value cache (KV cache) constraints. MFA enhances model capacity by efficiently scaling up both the number and dimension of attention heads through low-rank matrix factorization in the Query-Key (QK) circuit. Extending MFA, MFA-KR fur...| papers.cool
The self-attention mechanism traditionally relies on the softmax operator, necessitating positional embeddings like RoPE, or position biases to account for token order. But current methods using still face length generalisation challenges. We propose an alternative attention mechanism based on the stick-breaking process: For each token before the current, we determine a break point $\beta_{i,j}$, which represents the proportion of the remaining stick to allocate to the current token. We repea...| papers.cool
Multi-query attention (MQA), which only uses a single key-value head, drastically speeds up decoder inference. However, MQA can lead to quality degradation, and moreover it may not be desirable to train a separate model just for faster inference. We (1) propose a recipe for uptraining existing multi-head language model checkpoints into models with MQA using 5% of original pre-training compute, and (2) introduce grouped-query attention (GQA), a generalization of multi-query attention which use...| papers.cool
Multi-head attention layers, as used in the Transformer neural sequence model, are a powerful alternative to RNNs for moving information across and between sequences. While training these layers is generally fast and simple, due to parallelizability across the length of the sequence, incremental inference (where such paralleization is impossible) is often slow, due to the memory-bandwidth cost of repeatedly loading the large "keys" and "values" tensors. We propose a variant called multi-query...| papers.cool
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...| kexue.fm
自从DeepSeek爆火后，它所提的Attention变体MLA（Multi-head Latent Attention）也愈发受到关注。MLA通过巧妙的设计实现了MHA与MQA的自由切换，使得...| kexue.fm
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（$\boldsymbol{Q...| kexue.fm
我们知道，在RoPE中频率的计算公式为$\theta_i = b^{-2i/d}$，底数$b$默认值为10000。目前Long Context的主流做法之一是，先在$b=10000$上用短文本预...| kexue.fm
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...| kexue.fm
在文章《Transformer升级之路：20、MLA好在哪里?（上）》中，我们对MLA相比常见MHA、GQA、MQA的一些变化分别做了消融实验，其中的变化包括“增大head_dims”、“Par...| kexue.fm
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。...| kexue.fm