Skip to content
AI News HubLIVE
Original source2 min read

Geometric and Behavioral Stratification in Transformer Residual Streams

Summary

A new study finds that the prediction direction—the unembedding direction of the token a model currently predicts—acts as a content-defined privileged anchor in the residual stream. Anchored to this direction, residual-stream variation is geometrically and behaviorally stratified by proximity to the prediction: a narrow, scale-invariant prediction interface concentrates readout-relevant structure, while the vast prediction-distal complement expands with model scale. This stratification holds across 18 models (7B–120B, dense and MoE, base and instruction-tuned). Because the prediction direction is nearly orthogonal to principal variance axes, variance-based analyses recover this organization only partly, and the shortfall grows with prompt heterogeneity. Disrupting prediction-proximal vari…

SourcearXiv Machine LearningAuthor: Nelson Guda
Geometric and Behavioral Stratification in Transformer Residual Streams
Report an error

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

Correction instructions
Read article

[Submitted on 12 Aug 2026]

Title:Geometric and Behavioral Stratification in Transformer Residual Streams

View a PDF of the paper titled Geometric and Behavioral Stratification in Transformer Residual Streams, by Nelson Guda

View PDF HTML (experimental)

Abstract:Trained transformer models develop privileged bases: coordinate axes whose statistics differ from the rest of the residual stream. But what kind of direction does such a basis select? We investigate the prediction direction, the unembedding direction of the token a model currently predicts, and find that it functions as a content-defined privileged anchor. Measured with respect to this anchor, residual-stream variation is geometrically and behaviorally stratified by proximity to the prediction.

The stratification holds in all eighteen models tested (dense and mixture-of-experts, 7B-120B, base and instruction-tuned). A narrow, scale-invariant prediction interface concentrates readout-relevant structure, while the vast prediction-distal complement expands with model scale. Because the prediction direction sits nearly orthogonal to the principal variance axes, variance-based analyses recover this organization only partly, and the shortfall grows with prompt heterogeneity.

Anchoring reveals a steep geometric gradient: prediction-proximal regions are highly structured and cluster related prompts, while the complement is flatter and anti-discriminates among prompt groups. The interface is a narrow slice but functionally decisive. Disrupting the variance directions closest to the prediction causes immediate divergence and frequent task-frame shifts; disrupting the next level down delays divergence and preserves framing. The complement is weakly readout-aligned per direction yet causally and temporally load-bearing, and behavior is driven by direction rather than magnitude.

These results establish the prediction direction as a privileged anchor distinct from previously described coordinate axes, and give a geometric account of how high-dimensional computation coexists with linear readout.

Comments: 63 pages, 10 figures, 15 tables. Code and data: this https URL

Subjects:

Machine Learning (cs.LG); Computation and Language (cs.CL)

MSC classes: I.2.6, I.2.7

Cite as: arXiv:2608.12447 [cs.LG]

(or arXiv:2608.12447v1 [cs.LG] for this version)

https://doi.org/10.48550/arXiv.2608.12447

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Nelson Guda [view email] [v1] Wed, 12 Aug 2026 17:42:20 UTC (15,979 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Geometric and Behavioral Stratification in Transformer Residual Streams, by Nelson Guda

View PDF

HTML (experimental)

TeX Source

view license

Current browse context:

cs.LG

new | recent | 2026-08

Change to browse by:

cs cs.CL

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 prediction direction (unembedding direction of the currently predicted token) serves as a content-defined privileged anchor, producing geometric and behavioral stratification in the residual stream.
  • Stratification holds in all 18 tested models (7B–120B, dense and MoE, base and instruction-tuned); a narrow, scale-invariant prediction interface concentrates readout structure while the prediction-distal complement expands with scale.
  • The prediction direction lies nearly orthogonal to principal variance axes, so variance-based analyses capture only part of this organization, with the gap growing as prompt heterogeneity increases.
  • Disrupting prediction-proximal variance directions causes immediate divergence and frequent task-frame shifts; the complement is causally and temporally load-bearing, and behavior is driven by direction rather than magnitude.

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