mlsanityData to 7 Oct 2026

DecodeShare: Tracing the Shared Subspace of LLM Decode-Time Decisions

arXiv 2607.20469 ↗First posted 21 May 2026cs.AIcited by 1

On the map: Efficient LLM inference and serving → Early exit and edge inference

Abstract (arXiv)

Large language models (LLMs) handle many tasks with one set of parameters, but under KV-cached inference it is unclear what task-general structure, if any, is used at decode time rather than during prefill. We propose DecodeShare, a protocol that identifies a low-dimensional subspace consistently shared across tasks in decode-time hidden states, and then tests its causal role by removing that subspace only during decoding. In our experiments, disturbing the discovered shared subspace degrades decision performance far more than disturbing either a prefill-derived or random subspace under the same intervention budget. We further show this decode-shared subspace has practical consequences for activation steering: common steering directions can overlap the task-general decode channel. Projecting out this shared subspace directly separates the functional roles of the two components, while evaluating steering vectors at decode-time yields more reliable signal for downstream deployment than prefill-based proxies. Despite its compactness, the shared subspace can serve as a high-leverage causal channel at decode time. Code is available at: https://github.com/Zishan-Shao/decodeshare.git.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in Early exit and edge inference, one of 12 directions in Efficient LLM inference and serving. Builds on 40 of its 47 references that are in the map. First posted on arXiv 21 May 2026.

Builds on

40 of 47 references.

References · 40

  1. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  2. 2022Locating and Editing Factual Associations in GPTcited by 2,609
  3. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  4. 2019Similarity of Neural Network Representations Revisitedcited by 1,695
  5. 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,239
  6. 2022Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 smallcited by 1,130
  7. 2023Steering Llama 2 via Contrastive Activation Additioncited by 1,009
  8. 2024Refusal in Language Models Is Mediated by a Single Directioncited by 960
  9. 2022In-context Learning and Induction Headscited by 930
  10. 2024SnapKV: LLM Knows What You are Looking for Before Generationcited by 848
  11. 2023Steering Language Models With Activation Engineeringcited by 824
  12. 2023H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Modelscited by 805
28 more
  1. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 579
  2. 2022Transformer Feed-Forward Layers Build Predictions by Promoting Concepts in the Vocabulary Spacecited by 522
  3. 2023Model Tells You What to Discard: Adaptive KV Cache Compression for LLMscited by 466
  4. 2024PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funnelingcited by 420
  5. 2024Mechanistic Interpretability for AI Safety -- A Reviewcited by 415
  6. 2021Causal Abstractions of Neural Networkscited by 401
  7. 2023Towards Best Practices of Activation Patching in Language Models: Metrics and Methodscited by 328
  8. 2023Function Vectors in Large Language Modelscited by 292
  9. 2023Finding Alignments Between Interpretable Causal Variables and Distributed Neural Representationscited by 197
  10. 2020Causal Mediation Analysis for Interpreting Neural NLP: The Case of Gender Biascited by 183
  11. 2024Analyzing the Generalization and Reliability of Steering Vectorscited by 130
  12. 2023Circuit Component Reuse Across Tasks in Transformer Language Modelscited by 105
  13. 2025KVzip: Query-Agnostic KV Cache Compression with Context Reconstructioncited by 89
  14. 2023Is This the Subspace You Are Looking for? An Interpretability Illusion for Subspace Activation Patchingcited by 75
  15. 2024Eigen Attention: Attention in Low-Rank Space for KV Cache Compressioncited by 71
  16. 2024Finding Transformer Circuits with Edge Pruningcited by 58
  17. 2024Style Vectors for Steering Generative Large Language Modelcited by 56
  18. 2025The Universal Weight Subspace Hypothesiscited by 22
  19. 2024Discovering Low-rank Subspaces for Language-agnostic Multilingual Representationscited by 18
  20. 2025Unraveling LoRA Interference: Orthogonal Subspaces for Robust Model Mergingcited by 18
  21. 2025Adaptive Multi-Subspace Representation Steering for Attribute Alignment in Large Language Modelscited by 16
  22. 2025KV Cache Steering for Controlling Frozen LLMscited by 8
  23. 2025FlashSVD: Memory-Efficient Inference with Streaming for Low-Rank Modelscited by 8
  24. 2025Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behaviorcited by 7
  25. 2025Semantic Convergence: Investigating Shared Representations Across Scaled LLMscited by 3
  26. 2023Improving generalization in large language models by learning prefix subspacescited by 2
  27. 2025QKV Projections Require a Fraction of Their Memorycited by 2
  28. 2026FlashSVD v1.5: Making Low-Rank Transformers Inference Actually Fastcited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026What Does Activation Steering Control? Attribution Across Answer Encodings and Output-Sensitive Subspacescited by 1

Similar, not cited

Close in wording, no citation either way

  1. 2026The Missing Piece in Pre-trained Model Evaluation: Reward-Guided Decoding Unlocks Task-Oriented Behavior Without Parameter Updates
  2. 2026Interpreting and Steering State-Space Models via Activation Subspace Bottlenecks
  3. 2025Distribution-Aligned Decoding for Efficient LLM Task Adaptation
  4. 2026Where to Steer: Input-Dependent Layer Selection for Steering Improves LLM Alignment
  5. 2026Spillover-Aware Multi-Value Steering for Pluralistic LLM Alignment
  6. 2025Internal Chain-of-Thought: Empirical Evidence for Layer-wise Subtask Scheduling in LLMs
  7. 2026Memory Inception: Latent-Space KV Cache Manipulation for Steering LLMs
  8. 2024Controlling Large Language Model Agents with Entropic Activation Steering