With the development of large language models (LLMs), efficient inference through Key-Value (KV) cache compression has attracted considerable attention, especially for long-context generation. To compress the KV cache, recent methods identify critical KV tokens through static modeling of attention scores. However, these methods often struggle to accurately determine critical tokens as they neglect the temporal patterns in attention scores, resulting in a noticeable degradation in LLM performance. To address this challenge, we propose AttentionPredictor, which is the first learning-based method to directly predict attention patterns for KV cache compression and critical token identification. Specifically, AttentionPredictor learns a lightweight, unified convolution model to dynamically capture spatiotemporal patterns and predict the next-token attention scores. An appealing feature of AttentionPredictor is that it accurately predicts the attention score and shares the unified prediction model, which consumes negligible memory, among all transformer layers. Moreover, we propose a cross-token critical cache prefetching framework that hides the token estimation time overhead to accelerate the decoding stage. By retaining most of the attention information, AttentionPredictor achieves 13 KV cache compression and 5.6 speedup in a cache offloading scenario with comparable LLM performance, significantly outperforming the stateof-the-arts.

Due to the size and complexity of modern large language models (LLMs), it has proven challenging to uncover the underlying mechanisms that models use to solve reasoning problems. For instance, is their reasoning for a specific problem localized to certain parts of the network? Do they break down the reasoning problem into modular components that are then executed as sequential steps as we go deeper in the model? To better understand the reasoning capability of LLMs, we study a minimal propositional logic problem that requires combining multiple facts to arrive at a solution. By studying this problem on Mistral and Gemma models, up to 27B parameters, we illuminate the core components the models use to solve such logic problems. From a mechanistic interpretability point of view, we use causal mediation analysis to uncover the pathways and components of the LLMs' reasoning processes. Then, we offer fine-grained insights into the functions of attention heads in different layers. We not only find a sparse circuit that computes the answer, but we decompose it into sub-circuits that have four distinct and modular uses. Finally, we reveal that three distinct models - Mistral-7B, Gemma2-9B and Gemma-2-27B - contain analogous but not identical mechanisms.

Large language models are increasingly capable of handling long-context inputs, but the memory overhead of key-value (KV) cache remains a major bottleneck for general-purpose deployment. While various compression strategies have been explored, sequence-level compression, which drops the full KV caches for certain tokens, is particularly challenging as it can lead to the loss of important contextual information. To address this, we introduce UniGist, a sequence-level long-context compression framework that efficiently preserves context information by replacing raw tokens with special compression tokens (gists) in a fine-grained manner. We adopt a chunk-free training strategy and design an efficient kernel with a gist shift trick, enabling optimized GPU training. Our scheme also supports flexible inference by allowing the actual removal of compressed tokens, resulting in realtime memory savings. Experiments across multiple long-context tasks demonstrate that UniGist significantly improves compression quality, with especially strong performance in detail-recalling tasks and long-range dependency modeling.

In visual generation, the quadratic complexity of attention mechanisms results in high memory and computational costs, especially for longer token sequences required in high-resolution image or multi-frame video generation. To address this, prior research has explored techniques such as sparsification and quantization. However, these techniques face significant challenges under low density and reduced bitwidths. Through systematic analysis, we identify that the core difficulty stems from the dispersed and irregular characteristics of visual attention patterns. Therefore, instead of introducing specialized sparsification and quantization design to accommodate such patterns, we propose an alternative strategy: reorganizing the attention pattern to alleviate the challenges.

This paper studies the role of sinks and diagonal patterns as attention switch and anti-oversmoothing mechanisms. We analyze geometric conditions under which sinks can be represented, showing a necessary alignment between the embedding of the sink and all other embeddings. Next, we refine the current understanding of the role of sinks in oversmoothing prevention: we specify the conditions under which dense attention provably smooths more than sparse attention, and empirically verify that such conditions are often satisfied in practice. We further prove an equivalence between sinks and hard attention switch, in which the output of the attention is identically 0. Finally, we relax the hard attention switch by allowing token self-communication: we provide a quantitative comparison of the costs of representing sinks vs.\ diagonal patterns, showing why sinks are favored in pretrained transformers. The introduction and analysis of diagonal patterns and the generalization of the attention switch close the gap between what oversmoothing prevention requires and what sinks provide, while also establishing when and why attention layers act like MLPs if token communication is not necessary.

Conventional wisdom suggests that pre-training Vision Transformers (ViT) improves downstream performance by learning useful representations.

Neighborhood attention reduces the cost of self attention by restricting each token's attention span to its nearest neighbors. This restriction, parameterized by a window size and dilation factor, draws a spectrum of possible attention patterns between linear projection and self attention. Neighborhood attention, and more generally sliding window attention patterns, have long been bounded by infrastructure, particularly in higher-rank spaces (2-D and 3-D), calling for the development of custom kernels, which have been limited in either functionality, or performance, if not both. In this work, we aim to massively improve upon existing infrastructure by providing two new methods for implementing neighborhood attention. We first show that neighborhood attention can be represented as a batched GEMM problem, similar to standard attention, and implement it for 1-D and 2-D neighborhood attention.

Understanding the distance between human languages is central to linguistics, anthropology, and tracing human evolutionary history. Yet, while linguistics has long provided rich qualitative accounts of cross-linguistic variation, a unified and scalable quantitative approach to measuring language distance remains lacking. In this paper, we introduce a method that leverages pretrained multilingual language models as systematic instruments for linguistic measurement. Specifically, we show that the spontaneously emerged attention mechanisms of these models provide a robust, tokenization-agnostic measure of cross-linguistic distance, termed Attention Transport Distance (ATD). By treating attention matrices as probability distributions and measuring their geometric divergence via optimal transport, we quantify the representational distance between languages during translation. Applying ATD to a large and diverse set of languages, we demonstrate that the resulting distances recover established linguistic groupings with high fidelity and reveal patterns aligned with geographic and contact-induced relationships. Furthermore, incorporating ATD as a regularizer improves transfer performance in low-resource machine translation. Our results establish a principled foundation for testing linguistic hypotheses using artificial neural networks. This framework transforms multilingual models into powerful tools for quantitative linguistic discovery, facilitating more equitable multilingual AI.

Attention-based transformers have been remarkably successful at modeling generative processes across various domains and modalities.