Paper Detail

Maglev: Sliding Recurrent Memory

Bo Liu, Qiang Liu

huggingface Score 8.4

Published 2026-08-05 · First seen 2026-08-15

General AI

Abstract

We introduce , a recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention while remaining parallelizable during training. consists of two coupled models: a prefiller Q, which leverages full attentionIn practice, we use interleaved full and sliding-window attention for Q, as this yields stronger performance. The essential requirement is that Q be more expressive than P, with access to the full history. to produce memory targets m'_t, and a decoder P, which uses only sliding-window attention and recurrent K/V injection to produce decoder memories m_t for next-token prediction. We train with a memory consistency loss that aligns m_t with m'_t, allowing inference to use P alone. Empirically, improves validation loss and downstream pretraining benchmarks over sliding-window and latent recurrent transformer baselines. Moreover, sharing parameters between P and Q reduces parameter memory while preserving most of the gains.

Workflow Status

Review status
pending
Role
unreviewed
Read priority
soon
Vote
Not set.
Saved
no
Collections
Not filed yet.
Next action
Not filled yet.

Reading Brief

No structured notes yet. Add `summary_sections`, `why_relevant`, `claim_impact`, or `next_action` in `papers.jsonl` to enrich this view.

Why It Surfaced

No ranking explanation is available yet.

Tags

No tags.

BibTeX

@misc{liu2026maglev,
  title = {Maglev: Sliding Recurrent Memory},
  author = {Bo Liu and Qiang Liu},
  year = {2026},
  abstract = {We introduce , a recurrent Transformer architecture with fixed-size memory that generalizes sliding-window attention while remaining parallelizable during training. consists of two coupled models: a prefiller Q, which leverages full attentionIn practice, we use interleaved full and sliding-window attention for Q, as this yields stronger performance. The essential requirement is that Q be more expressive than P, with access to the full history. to produce memory targets m'\_t, and a decoder P, whi},
  url = {https://huggingface.co/papers/2608.02870},
  keywords = {recurrent Transformer, sliding-window attention, full attention, memory consistency loss, recurrent K/V injection, prefiller, decoder, parameter sharing, huggingface daily},
  eprint = {2608.02870},
  archiveprefix = {arXiv},
}

Metadata

{}