Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity

Hugging Face Daily Papers Papers

Summary

Sparse Delta Memory extends gated linear RNNs with sparse addressing to dramatically increase hidden state capacity for improved long-context learning and retrieval while maintaining computational efficiency.

Linear attention models allow a fixed state size and a fixed amount of compute per token. However, due to their limited state size, linear attention models fall behind in long-context recall compared to softmax-attention-based transformer architectures. Increasing the state size of linear attention improves recall performance but at the cost of higher FLOPs. In this work, we introduce Sparse Delta Memory (SDM), an architecture that scales the hidden state of gated linear RNNs to orders of magnitude higher capacity using a sparse addressing scheme. SDM extends the Gated DeltaNet architecture by replacing the dense key-value outer product with sparse reads and writes to a large explicit memory. We show that, under an isoFLOP constraint and with an identical number of parameters, a higher state memory capacity significantly improves performance on in-context learning and long-context retrieval tasks. Moreover, by learning the initial state of the SDM memory and therefore using it as a parametric memory, we show that the model further improves on a wide range of common-knowledge and reasoning tasks.
Original Article
View Cached Full Text

Cached at: 07/09/26, 03:40 PM

Paper page - Sparse Delta Memory: Scaling the State of Linear RNNs through Sparsity

Source: https://huggingface.co/papers/2607.07386

Abstract

Sparse Delta Memory extends gated linear RNNs with sparse addressing to dramatically increase hidden state capacity for improved long-context learning and retrieval while maintaining computational efficiency.

Linear attention modelsallow a fixed state size and a fixed amount of compute per token. However, due to their limited state size,linear attention modelsfall behind in long-context recall compared tosoftmax-attention-basedtransformer architectures. Increasing the state size of linear attention improves recall performance but at the cost of higher FLOPs. In this work, we introduce Sparse Delta Memory (SDM), an architecture that scales the hidden state ofgated linear RNNsto orders of magnitude higher capacity using asparse addressing scheme. SDM extends theGated DeltaNetarchitecture by replacing the dense key-value outer product with sparse reads and writes to a largeexplicit memory. We show that, under an isoFLOP constraint and with an identical number of parameters, a higher state memory capacity significantly improves performance onin-context learningandlong-context retrievaltasks. Moreover, by learning the initial state of the SDM memory and therefore using it as aparametric memory, we show that the model further improves on a wide range of common-knowledge and reasoning tasks.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2607\.07386

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2607.07386 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2607.07386 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2607.07386 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Δ-Mem: Efficient Online Memory for Large Language Models

Hacker News Top

Proposes delta-Mem, a lightweight online memory mechanism that uses a compact state matrix updated by delta-rule learning to improve long-context performance of frozen LLMs without full fine-tuning or context extension.

Dynamic Linear Attention

Hugging Face Daily Papers

DLA introduces adaptive state merging and capacity-bounded memory modeling for multi-state linear attention, improving long-context LLM performance.

Dynamic Linear Attention

arXiv cs.CL

This paper proposes DLA, a dynamic memory modeling framework for multi-state linear attention that adaptively merges states based on token information variation and maintains a fixed-size state cache, enabling better long-context representation without the quadratic complexity of standard attention.