◇ 今日论文 · Maglev: Sliding Recurrent Memory
① 这是什么
今日论文《今日论文 · Maglev: Sliding Recurrent Memory》（Bo Liu、Qiang Liu），来自 HF Daily Papers 公开源。
② 值不值得用
进入 HF 每日精选说明有社区关注度；是否与你的问题相关需要读原文判断，本条目不评价研究质量。
③ 怎么开始
先读摘要，需要再读全文（https://huggingface.co/papers/2608.02870）；对照论文 ID（2608.02870）可找实现与讨论。
④ 关键证据
- 论文 ID 2608.02870
- 作者 Bo Liu、Qiang Liu
⑤ 注意事项与坑
摘要来自作者原文，未做同行评议级核验；引用请以正式发表版本为准。
相关推荐（反茧房 · 应该知道的）
◇ 今日论文 · RibAssist 3D: Biplanar Rib-Fracture Detection, Addressing, and Selective 3D Localization from CT-Derived Projections
Rib fractures are common and time-consuming to localize on computed tomography (CT). We ask whether fractures detected independently in two orthogonal CT-derived projections (anteroposterior and lateral) can be paired across views and triangulated into reliable 3D points at a controlled rate of false outputs, and we answer it with a staged diagnostic study. The projection geometry is exact, and gi
相似◇ 今日论文 · Thought-Level Beam Search for Reasoning
Test-time compute scaling is a primary driver of performance in large reasoning models (LRMs), but extreme inefficiency bounds current approaches, shifting the critical question from how much compute to spend, to where to allocate it. We formalize test-time reasoning as a constrained compute allocation problem over partial trajectories. Under a fixed hardware budget, existing paradigms fail to act
相似◇ 今日论文 · Mitigating Gender Bias in English to Romanian Machine Translation
Machine translation (MT) systems often fail to correctly translate gender, especially when converting from a gender-neutral language like English to a gendered target language such as Romanian. This bias results in translations that default to masculine forms or reinforce gender stereotypes. We propose a hybrid pipeline to mitigate this issue by combining large language model (LLM)-based gender cl
这条不对 / 我执行报错了？带 entry_id 一键提 Issue，回流后修订版本。
🐞 纠错 / 报错 → 提 Issue{
"schema_version": "0.3",
"entry_id": "ASR-PAPER-20260816-2608-02870",
"record_type": "paper",
"title": "今日论文 · Maglev: Sliding Recurrent Memory",
"observed_at": "2026-08-16",
"source": {
"name": "huggingface.co/api/daily_papers",
"url": "https://huggingface.co/api/daily_papers?limit=10",
"snapshot_hash": "735b0882f0ad228e42c868385e901b1b1e78b57fb562534343256eb58fd33f9a",
"fetched_at": "2026-08-16T10:45:59+00:00"
},
"confidence": "medium",
"human_stream": {
"summary": "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 exp",
"key_points": [
"论文 ID 2608.02870",
"作者 Bo Liu、Qiang Liu",
"发布 2026-08-05T00:00:00.000Z"
],
"note": "来源为 HuggingFace daily papers 公开 API；完整阅读请走原文链接。",
"what_it_is": "今日论文《今日论文 · Maglev: Sliding Recurrent Memory》（Bo Liu、Qiang Liu），来自 HF Daily Papers 公开源。",
"worth_it": "进入 HF 每日精选说明有社区关注度；是否与你的问题相关需要读原文判断，本条目不评价研究质量。",
"how_to_start": "先读摘要，需要再读全文（https://huggingface.co/papers/2608.02870）；对照论文 ID（2608.02870）可找实现与讨论。",
"evidence": [
"论文 ID 2608.02870",
"作者 Bo Liu、Qiang Liu"
],
"cautions": "摘要来自作者原文，未做同行评议级核验；引用请以正式发表版本为准。"
},
"ai_stream": {
"structured": {
"paper_id": "2608.02870",
"title": "Maglev: Sliding Recurrent Memory",
"authors": [
"Bo Liu",
"Qiang Liu"
],
"published_at": "2026-08-05T00:00:00.000Z",
"url": "https://huggingface.co/papers/2608.02870",
"source_label": "hf-daily-papers"
},
"raw": [
{
"paper_id": "2608.02870",
"title": "Maglev: Sliding Recurrent Memory",
"authors": [
"Bo Liu",
"Qiang Liu"
],
"summary": "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 f",
"published_at": "2026-08-05T00:00:00.000Z",
"submitted_on_daily_at": null,
"url": "https://huggingface.co/papers/2608.02870"
}
]
},
"token_cost": {
"total": 0.0,
"currency": "USD",
"breakdown": {
"crawl": 0.0,
"clean": 0.0,
"elevate": 0.0,
"verify": 0.0
}
},
"machine_verified": true,
"review": {
"status": "agent_reviewed",
"reviewer": "pipeline-validate",
"reviewed_at": "2026-08-16T12:12:04+00:00",
"comments": "机检通过（schema 0 error + 溯源一致 + 成本达标）"
},
"provenance": {
"extracted_by": "codex",
"extracted_at": "2026-08-16T10:45:59+00:00",
"pipeline": "collect_hf_papers.py + build_entries.py v0.2（规则管线）",
"access_urls": [
"https://huggingface.co/api/daily_papers?limit=10"
],
"card_generated_at": "2026-08-16T12:12:03+00:00",
"card_pipeline": "enrich_human_stream.py v0.1（规则模板，零 Token）"
}
}