mlsanityData to 7 Oct 2026

Latent Space Communication via K-V Cache Alignment

arXiv 2601.06123 ↗First posted 4 Jan 2026cs.LG cs.AIcited by 9

Not on the map. Why?

Abstract (arXiv)

Solving increasingly complex problems with large language models (LLMs) necessitates a move beyond individual models and towards multi-model systems that can effectively collaborate. While text has traditionally served as the medium for inter-model communication, a richer and more efficient exchange is possible if models can access each other's internal states directly. In this paper, we propose learning a shared representation space that aligns the k-v caches of multiple models, creating a high-bandwidth channel for collaboration without altering the underlying pre-trained parameters. We do so by augmenting each model with adapters to translate its state into and out of this shared space. Via a suite of experiments with Gemma-2 models, we demonstrate that this approach not only enables seamless inter-model communication but also improves individual model performance. We also show that the shared space allows for the direct transfer of learned skills, such as soft prompts, between different models. Our work represents a significant step towards a future where models can fluidly share knowledge and capabilities.

Key facts. Cited by 9 papers in the map (data through 7 Oct 2026). Builds on 32 of its 42 references that are in the map. First posted on arXiv 4 Jan 2026.

Builds on

32 of 42 references.

References · 32

  1. 2017Attention Is All You Needcited by 59,932
  2. 2020Language Models are Few-Shot Learnerscited by 32,199
  3. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,491
  4. 2016Layer Normalizationcited by 5,579
  5. 2022PaLM: Scaling Language Modeling with Pathwayscited by 5,304
  6. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 4,993
  7. 2021Prefix-Tuning: Optimizing Continuous Prompts for Generationcited by 3,878
  8. 2021The Power of Scale for Parameter-Efficient Prompt Tuningcited by 3,769
  9. 2021Finetuned Language Models Are Zero-Shot Learnerscited by 3,747
  10. 2016Gaussian Error Linear Units (GELUs)cited by 3,431
  11. 2022Training Compute-Optimal Large Language Modelscited by 2,927
  12. 2016Deep Learning with Differential Privacycited by 2,682
20 more
  1. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,928
  2. 2024The AI Scientist: Towards Fully Automated Open-Ended Scientific Discoverycited by 1,002
  3. 2024OpenHands: An Open Platform for AI Software Developers as Generalist Agentscited by 956
  4. 2024Training Large Language Models to Reason in a Continuous Latent Spacecited by 763
  5. 2016Learning to Communicate with Deep Multi-Agent Reinforcement Learningcited by 561
  6. 2022Git Re-Basin: Merging Models modulo Permutation Symmetriescited by 460
  7. 2020Do Wide and Deep Networks Learn the Same Things? Uncovering How Neural Network Representations Vary with Width and Depthcited by 241
  8. 2024Evolutionary Optimization of Model Merging Recipescited by 204
  9. 2022Relative representations enable zero-shot latent space communicationcited by 175
  10. 2020Emergent Multi-Agent Communication in the Deep Learning Eracited by 168
  11. 2023Large Language Model Cascades with Mixture of Thoughts Representations for Cost-efficient Reasoningcited by 140
  12. 2025Harnessing the Universal Geometry of Embeddingscited by 54
  13. 2023Latent Space Translation via Semantic Alignmentcited by 46
  14. 2023Interactive Natural Language Processingcited by 31
  15. 2023Git-Theta: A Git Extension for Collaborative Development of Machine Learning Modelscited by 15
  16. 2024DiPaCo: Distributed Path Compositioncited by 9
  17. 2024No Need to Talk: Asynchronous Mixture of Language Modelscited by 5
  18. 2023Cheap Talk Discovery and Utilization in Multi-Agent Reinforcement Learningcited by 4
  19. 2024Assessing the Portability of Parameter Matrices Trained by Parameter-Efficient Finetuning Methodscited by 2
  20. 2025Language Model Networks: Supervision-Efficient Learning through Dense Communicationcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 9.

9 papers.

  1. 2026Cross-Model KV Cache Transfer in LLM Families: A Closed-Form Linear Mapping for Prefill Reusecited by 10
  2. 2026Latent Cache Flow: Model-to-Model Communication Without Textcited by 6
  3. 2026KV Cache Translation across Heterogeneous Large Language Modelscited by 4
  4. 2026When Latent Agents Lie: KV-Cache Integrity in Multi-Agent LLM Collaborationcited by 2
  5. 2026CacheBridge: Efficient Cross-Model KV Cache Transfercited by 2
  6. 2026LatentDx: Latent Multi-Agent Communication for Cross-Hospital Rare-Disease Diagnosiscited by 1
  7. 2026LACO: Adaptive Latent Communication for Collaborative Drivingcited by 0
  8. 2026PReCache: Efficient KV Cache Sharing for Multi-LoRA Agents via Low-Rank Precomputation and Neutral Reconstructioncited by 0
  9. 2026RaReCache: Bridging the Gap in Cross-Model KV Cache Reuse via Rank disagreement-based Selective Recomputationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Cache-to-Cache: Direct Semantic Communication Between Large Language Models
  2. 2026RelayCaching: Accelerating LLM Collaboration via Decoding KV Cache Reuse
  3. 2024Beyond KV Caching: Shared Attention for Efficient LLMs
  4. 2025Direct Semantic Communication Between Large Language Models via Vector Translation
  5. 2024Deliberation in Latent Space via Differentiable Cache Augmentation
  6. 2025IC-Cache: Efficient Large Language Model Serving via In-context Caching
  7. 2024A Systematic Study of Cross-Layer KV Sharing for Efficient LLM Inference
  8. 2026Towards Efficient Large Language Model Serving: A Survey on System-Aware KV Cache Optimization