analiza aiqbench
This research introduces Key-Value Means (KVM), a novel attention mechanism that unifies the expandable context of transformers with the efficiency of recurrent neural networks. By enabling fixed or growing state sizes without custom kernels, KVM addresses critical bottlenecks in long-context processing and memory scaling for large language models. The approach offers significant architectural flexibility, allowing practitioners to tune prefill time complexity between linear and quadratic regimes using standard operations.
Key-Value Means: Transformers with Expandable Block-Recurrent Compressed Memory
arXiv:2605.09877v4 Announce Type: replace-cross Abstract: Recall presents a difficult choice: transformers have a linearly growing memory that slows each successive token, while linear RNNs typically have fixed costs but limited recall. We present Key-Value Means ("KVM"), a novel block-recurrence for attention that can accommodate either fixed-size or growing state. Equipping a strong transformer baseline with fixed-size KVM attention layers yields a strong $O(N)$ chunked RNN, while adding only an insignificant number of new parameters. We train a transformer with a growable KVM cache and show it performs competitively on long-context tests with only subquadratic prefill time and sublinear state growth. KVM is implementable with standard operations and without custom kernels, and supports chunk-wise parallelizable training and prefill. It provides many of the benefits of both traditional transformers (expandable context memory, chunk-wise parallelizable training and prefill) and RNNs in a single unified package. It can be used on every layer, saving KV-cache memory, and allowing a continuous range of choices of prefill time complexity between $O(N)$ and $O(N^2)$. We release our code at https://github.com/featherless-ai/KVM-paper and trained models at https://huggingface.co/collections/featherless-ai/kvm-paper under the Apache 2.0 license.
Abstract:Recall presents a difficult choice: transformers have a linearly growing memory that slows each successive token, while linear RNNs typically have fixed costs but limited recall. We present Key-Value Means ("KVM"), a novel block-recurrence for attention that can accommodate either fixed-size or growing state. Equipping a strong transformer baseline with fixed-size KVM attention layers yields a strong $O(N)$ chunked RNN, while adding only an insignificant number of new parameters. We train a transformer with a growable KVM cache and show it performs competitively on long-context tests with only subquadratic prefill time and sublinear state growth. KVM is implementable with standard operations and without custom kernels, and supports chunk-wise parallelizable training and prefill. It provides many of the benefits of both traditional transformers (expandable context memory, chunk-wise parallelizable training and prefill) and RNNs in a single unified package. It can be used on every layer, saving KV-cache memory, and allowing a continuous range of choices of prefill time complexity between $O(N)$ and $O(N^2)$. We release our code at this https URL and trained models at this https URL under the Apache 2.0 license.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL)
arXiv:2605.09877 [cs.LG]
(or arXiv:2605.09877v4 [cs.LG] for this version)
Text integral sub licența deschisă a publicației · 200 cuvinte
Subiecte conexe
Productionizing TurboQuant on AMD GPUs for KV-Cache-Bound LLM Inference
AMD · 11 iun. 2026
Notes on Algorithms
LessWrong · 10 iun. 2026
How do neural networks regress cubic polynomials? Apparently, they use a trick invented in Milan 500 years ago
LessWrong · 22 iul. 2026
DI Series: Serving Kimi-K3 MXFP4 Wide-EP Disaggregated on AMD Instinct MI300X and MI325X
AMD · 8 oct. 2026