Hugging Face Daily Papers · · 3 min read

LongAttnComp: Cross-Family Context Compression for Long-Context Reasoning

Mirrored from Hugging Face Daily Papers for archival readability. Support the source by reading on the original site.

This paper presents a fine-tuning-based long-context compressor with a two-stage training recipe , which achieves strong long-context reasoning performance. A shorter version was accepted at the AdaptFM Workshop at ICML 2026.</p>\n","updatedAt":"2026-06-02T04:00:39.399Z","author":{"_id":"6813fbe428c1f6200bea5685","avatarUrl":"/avatars/60d3c8dfb85cad54c5ebc9f6e44100db.svg","fullname":"mengmeng ji","name":"mengmengj","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":1,"isUserFollowing":false}},"numEdits":1,"identifiedLanguage":{"language":"en","probability":0.9423032999038696},"editors":["mengmengj"],"editorAvatarUrls":["/avatars/60d3c8dfb85cad54c5ebc9f6e44100db.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2606.01336","authors":[{"_id":"6a1e52fd808ddbc3c7d43d4f","name":"Mengmeng Ji","hidden":false},{"_id":"6a1e52fd808ddbc3c7d43d50","name":"Ravi Shanker Raju","hidden":false},{"_id":"6a1e52fd808ddbc3c7d43d51","name":"Jonathan Lingjie Li","hidden":false},{"_id":"6a1e52fd808ddbc3c7d43d52","name":"Chen Wu","hidden":false}],"publishedAt":"2026-05-31T00:00:00.000Z","submittedOnDailyAt":"2026-06-02T00:00:00.000Z","title":"LongAttnComp: Cross-Family Context Compression for Long-Context Reasoning","submittedOnDailyBy":{"_id":"6813fbe428c1f6200bea5685","avatarUrl":"/avatars/60d3c8dfb85cad54c5ebc9f6e44100db.svg","isPro":false,"fullname":"mengmeng ji","user":"mengmengj","type":"user","name":"mengmengj"},"summary":"As real-world applications increasingly require processing inputs of 100k+ tokens, the gap between context length and inference efficiency has become a critical bottleneck. Context compression offers a way to reduce prefill costs while preserving task accuracy. However, existing training-free attention-based methods leave substantial gaps in demanding long-context tasks such as code reasoning. We present LongAttnComp, a long-context adaptation of AttnComp that fine-tunes a lightweight cross-attention scoring layer and introduces tokenlevel chunking, a token-budget top-p algorithm, positional reordering, and a formatagnostic query parser. We further design a two-stage fine-tuning recipe for the compressor: Stage 1 builds a general retrieval foundation from NIAH-style data, and Stage 2 extends it with multi-hop and reasoning data for broader long-context task coverage. On InfiniteBench Code-Debug, LongAttnComp matches or exceeds full-context accuracy, substantially outperforms training-free baselines, and transfers across four target models from three families. On LongBench v2, the two-stage recipe largely closes the Stage 1 gap on multi-document reasoning while preserving Code-Debug performance.","upvotes":2,"discussionId":"6a1e52fd808ddbc3c7d43d53","ai_summary":"LongAttnComp adapts AttnComp for long-context processing by fine-tuning lightweight attention layers and implementing token-level chunking and positional reordering techniques.","ai_keywords":["attention-based methods","context compression","prefill costs","token-level chunking","top-p algorithm","positional reordering","query parser","two-stage fine-tuning","retrieval foundation","multi-hop reasoning","long-context tasks"],"organization":{"_id":"645b3d8d87c79b6ec0b959eb","name":"sambanovasystems","fullname":"SambaNova","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d14661053a863f53737f40/MQKwU5Fmv64m1XM38Y7R9.png"}},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"6a15e80d9dc90b65ed374ac7","avatarUrl":"/avatars/2ede63c0ccfcfb575bb39e20774a7e45.svg","isPro":false,"fullname":"Лебедев Татьяна","user":"evelyncj9o","type":"user"},{"_id":"69ccf213409c56b8cbf25069","avatarUrl":"/avatars/a89d549641389e085dc4376bd052692d.svg","isPro":false,"fullname":"Григорьев Полина","user":"charlottesanc","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"organization":{"_id":"645b3d8d87c79b6ec0b959eb","name":"sambanovasystems","fullname":"SambaNova","avatar":"https://cdn-avatars.huggingface.co/v1/production/uploads/65d14661053a863f53737f40/MQKwU5Fmv64m1XM38Y7R9.png"},"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2606/2606.01336.md"}">
Papers
arxiv:2606.01336

LongAttnComp: Cross-Family Context Compression for Long-Context Reasoning

Published on May 31
· Submitted by
mengmeng ji
on Jun 2
Authors:
,
,
,

Abstract

LongAttnComp adapts AttnComp for long-context processing by fine-tuning lightweight attention layers and implementing token-level chunking and positional reordering techniques.

AI-generated summary

As real-world applications increasingly require processing inputs of 100k+ tokens, the gap between context length and inference efficiency has become a critical bottleneck. Context compression offers a way to reduce prefill costs while preserving task accuracy. However, existing training-free attention-based methods leave substantial gaps in demanding long-context tasks such as code reasoning. We present LongAttnComp, a long-context adaptation of AttnComp that fine-tunes a lightweight cross-attention scoring layer and introduces tokenlevel chunking, a token-budget top-p algorithm, positional reordering, and a formatagnostic query parser. We further design a two-stage fine-tuning recipe for the compressor: Stage 1 builds a general retrieval foundation from NIAH-style data, and Stage 2 extends it with multi-hop and reasoning data for broader long-context task coverage. On InfiniteBench Code-Debug, LongAttnComp matches or exceeds full-context accuracy, substantially outperforms training-free baselines, and transfers across four target models from three families. On LongBench v2, the two-stage recipe largely closes the Stage 1 gap on multi-document reasoning while preserving Code-Debug performance.

Community

This paper presents a fine-tuning-based long-context compressor with a two-stage training recipe , which achieves strong long-context reasoning performance. A shorter version was accepted at the AdaptFM Workshop at ICML 2026.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.
Tap or paste here to upload images

· Sign up or log in to comment

Get this paper in your agent:

hf papers read 2606.01336
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2606.01336 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2606.01336 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2606.01336 in a Space README.md to link it from this page.

Collections including this paper 0

No Collection including this paper

Add this paper to a collection to link it from this page.

Discussion (0)

Sign in to join the discussion. Free account, 30 seconds — email code or GitHub.

Sign in →

No comments yet. Sign in and be the first to say something.

More from Hugging Face Daily Papers