Skip to content
AI News HubLIVE
Original source2 min read

Models Take Notes at Prefill: KV Cache Can Be Editable and Composable

Summary

A study reveals that during prefill, LLMs write field-conditioned conclusions onto downstream 'notes', making KV cache editable and composable. Editing a field with chain-of-thought recovers decisions with ~1% compute, while precompiled skills can be position-portable and spliced into any context, achieving near-identical outputs to full recompute with up to 14.9x lower latency. The approach is validated across scales and cache types, achieving 98.5% prefix cache hit rate in production benchmarks.

SourcearXiv Machine LearningAuthor: Bojie Li
Models Take Notes at Prefill: KV Cache Can Be Editable and Composable
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[2606.17107] Models Take Notes at Prefill: KV Cache Can Be Editable and Composable

[Submitted on 14 Jun 2026]

Title:Models Take Notes at Prefill: KV Cache Can Be Editable and Composable

View a PDF of the paper titled Models Take Notes at Prefill: KV Cache Can Be Editable and Composable, by Bojie Li

View PDF HTML (experimental)

Abstract:Prefix caching reuses prefill only across an exactly shared prefix, so one changed field invalidates the entire downstream cache. Yet overwriting the field's own key/value vectors and reusing the rest leaves the model acting on the old value. The reason, established causally across four model families: at prefill the model has already written the field-conditioned conclusion onto downstream notes; the field's own key/value drives under 1% of the decision. Read as a notebook of memoized conclusions, two capabilities follow. (1) It is editable. A salient erratum amends the notes; and with chain-of-thought, editing the field alone recovers the decision (1.00 at 8B, ~1% compute), while without CoT it is ignored. (2) It is composable. The notes are position-portable, so a precompiled skill can be RoPE-repositioned and spliced into any context, indistinguishable from full recompute (logit cosine 0.90-0.999, twelve models) at O(L) rather than O(L^2) time-to-first-token. A unified edit+compose agent stays decision-identical to recompute at up to 14.9x lower latency. The approach applies to any per-token attention KV cache, validated across scale, quantization, Mixture-of-Experts, and multimodal caches, and extends to several attention variants through small adapters. Because the erratum is append-only, it composes with production prefix caching: in an online vLLM benchmark it keeps the prefix cache-aligned (98.5% hit-rate), cutting p90 time-to-first-token by 53-398x.

Subjects:

Machine Learning (cs.LG); Artificial Intelligence (cs.AI)

Cite as: arXiv:2606.17107 [cs.LG]

(or arXiv:2606.17107v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2606.17107

arXiv-issued DOI via DataCite

Submission history

From: Bojie Li [view email] [v1] Sun, 14 Jun 2026 15:31:23 UTC (534 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Models Take Notes at Prefill: KV Cache Can Be Editable and Composable, by Bojie Li

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.LG

new | recent | 2026-06

Change to browse by:

cs cs.AI

References & Citations

NASA ADS

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

IArxiv recommender toggle

IArxiv Recommender (What is IArxiv?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • The model's own key/value vectors contribute less than 1% to the final decision; the decision is stored in downstream notes written during prefill.
  • Editing a field with chain-of-thought corrects errors with perfect accuracy (1.00 at 8B) using only ~1% of compute.
  • Notes are position-portable, allowing precompiled skills to be repositioned and spliced into any context, indistinguishable from full recompute (logit cosine 0.90-0.999).
  • A unified edit+compose agent is decision-identical to recompute with up to 14.9x lower latency and composes with production prefix caching (98.5% hit rate).

Highlights and analysis are generated automatically and may contain errors. Check the original source.

More recent stories on this topic · Models