---
type: "post"
id: "proxyattn-a-training-free-method-from-researchers-at-59312c"
slug: "proxyattn-a-training-free-method-from-researchers-at-59312c"
title: "ProxyAttn , a training-free method from researchers at Harbin Institute of Technology and Baidu, selects those blocks using exact token-level scores from one or more representative attention heads."
description: "ProxyAttn tests whether attention-block selection can be made more precise without training a separate estimator."
retrieval_nugget: "ProxyAttn tests whether attention-block selection can be made more precise without training a separate estimator."
published_at: "2026-10-15"
updated_at: "2026-10-15"
record_date: "2026-10-15"
date_kind: "published_at"
topics: ["models","inference"]
source_urls: ["https://arxiv.org/abs/2509.24745"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:proxyattn-a-training-free-method-from-researchers-at-59312c"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:a22bdc48ed0633a46ef923bcfb607354d179d82cfacb253000d4aeca058ae726","reviewed_at":"2026-10-15T07:01:32.401Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"proxy-head scores can guide sparse attention while dynamic budgets adapt the number of selected blocks to each head.","observed_facts":[{"text":"The arXiv abstract describes ProxyAttn as a training-free sparse attention algorithm that uses pooled representative heads to approximate scores for all heads.","source_urls":["https://arxiv.org/abs/2509.24745"]},{"text":"The abstract reports up to 10.3× attention acceleration and 2.4× prefill acceleration without significant performance loss.","source_urls":["https://arxiv.org/abs/2509.24745"]},{"text":"The arXiv page identifies the paper as an ICLR 2026 camera-ready version and states that code is available.","source_urls":["https://arxiv.org/abs/2509.24745"]}],"mechanism":"representative heads provide a lower-cost signal for block importance, while head-specific budgets address variation in sparsity needs.","why_now":"long-context prefill remains costly because attention has quadratic complexity, and the paper positions dynamic block selection as a way to reduce that burden.","implications":["the method makes representative-head selection and per-head budget behavior important implementation and evaluation variables.","Builders should compare quality and end-to-end prefill latency rather than infer deployment gains from kernel speed alone."],"evidence_boundary":"these are maxima reported by the authors across their evaluations, not independent confirmation or a guarantee for every model, sequence length, hardware setup, or sparsity target.","watch_conditions":["whether the released implementation reproduces the reported speedups across models and realistic long-context workloads."],"related_records":[],"new_branch_reason":"No prior New Runtime analytical record was supplied; this creates a distinct record on representative-head-guided sparse attention."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/proxyattn-a-training-free-method-from-researchers-at-59312c/","markdown":"https://newruntime.com/posts/proxyattn-a-training-free-method-from-researchers-at-59312c.md","json":"https://newruntime.com/posts/proxyattn-a-training-free-method-from-researchers-at-59312c.json"}
---

# ProxyAttn , a training-free method from researchers at Harbin Institute of Technology and Baidu, selects those blocks using exact token-level scores from one or more representative attention heads.

## Retrieval answer

ProxyAttn tests whether attention-block selection can be made more precise without training a separate estimator.

ProxyAttn tests whether attention-block selection can be made more precise without training a separate estimator. **Thesis:** proxy-head scores can guide sparse attention while dynamic budgets adapt the number of selected blocks to each head. The paper describes pooling representative heads to approximate scores for other heads, then combining those estimates with multi-head dynamic budget estimation. **Mechanism:** representative heads provide a lower-cost signal for block importance, while head-specific budgets address variation in sparsity needs.

**Why now:** long-context prefill remains costly because attention has quadratic complexity, and the paper positions dynamic block selection as a way to reduce that burden. The arXiv page reports up to 10.3× attention acceleration and 2.4× prefill acceleration without significant performance loss; it also identifies the work as an ICLR 2026 camera-ready paper and says code is available. **Evidence boundary:** these are maxima reported by the authors across their evaluations, not independent confirmation or a guarantee for every model, sequence length, hardware setup, or sparsity target.

The practical implication is that builders evaluating sparse attention should compare quality and end-to-end prefill latency, not infer deployment gains from kernel speed alone. **First implication:** the method makes representative-head selection and per-head budget behavior important implementation and evaluation variables. **What to watch:** whether the released implementation reproduces the reported speedups across models and realistic long-context workloads. The evidence supports a research result, not a claim that ProxyAttn is already integrated into production systems.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)