mlsanityData to 14 Oct 2026

Your Transformer is Secretly Linear

arXiv 2405.12250 ↗First posted 19 May 2024cs.LG cs.AI cs.CLcited by 23

Not on the map. Why?

Abstract (arXiv)

This paper reveals a novel linear characteristic exclusive to transformer decoders, including models such as GPT, LLaMA, OPT, BLOOM and others. We analyze embedding transformations between sequential layers, uncovering a near-perfect linear relationship (Procrustes similarity score of 0.99). However, linearity decreases when the residual component is removed due to a consistently low output norm of the transformer layer. Our experiments show that removing or linearly approximating some of the most linear blocks of transformers does not affect significantly the loss or model performance. Moreover, in our pretraining experiments on smaller models we introduce a cosine-similarity-based regularization, aimed at reducing layer linearity. This regularization improves performance metrics on benchmarks like Tiny Stories and SuperGLUE and as well successfully decreases the linearity of the models. This study challenges the existing understanding of transformer architectures, suggesting that their operation may be more linear than previously assumed.

Key facts. Cited by 23 papers in the map (data through 14 Oct 2026). Builds on 20 of its 27 references that are in the map. First posted on arXiv 19 May 2024.

Builds on

20 of 27 references.

References · 20

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  3. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,720
  4. 2019SuperGLUE: A Stickier Benchmark for General-Purpose Language Understanding Systemscited by 1,729
  5. 2023SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shotcited by 1,121
  6. 2022Toy Models of Superpositioncited by 971
  7. 2020Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuningcited by 671
  8. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  9. 2023TinyStories: How Small Can Language Models Be and Still Speak Coherent English?cited by 449
  10. 2021A Survey of Transformerscited by 401
  11. 2024SliceGPT: Compress Large Language Models by Deleting Rows and Columnscited by 385
  12. 2019Intrinsic dimension of data representations in deep neural networkscited by 282
8 more
  1. 2023Jump to Conclusions: Short-Cutting Transformers With Linear Transformationscited by 90
  2. 2023A Comprehensive Survey on Applications of Transformers for Deep Learning Taskscited by 71
  3. 2023LoRAPrune: Structured Pruning Meets Low-Rank Parameter-Efficient Fine-Tuningcited by 53
  4. 2016Pruning Convolutional Neural Networks for Resource Efficient Inferencecited by 48
  5. 2024Everybody Prune Now: Structured Pruning of LLMs with only Forward Passescited by 43
  6. 2023The Shape of Learning: Anisotropy and Intrinsic Dimensions in Transformer-Based Modelscited by 39
  7. 2023BabyStories: Can Reinforcement Learning Teach Baby Language Models to Write Better Stories?cited by 7
  8. 2024MetaOptimize: A Framework for Optimizing Step Sizes and Other Meta-parameterscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2024: 2; 2025: 11; 2026: 10.

23 papers.

Sort:

By area:

  1. 2025Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coincited by 52
  2. 2025Cache Me If You Must: Adaptive Key-Value Quantization for Large Language Modelscited by 24
  3. 2024AdaptAgent: Adapting Multimodal Web Agents with Few-Shot Learning from Human Demonstrationscited by 19
  4. 2025ReplaceMe: Network Simplification via Depth Pruning and Transformer Block Linearizationcited by 13
  5. 2025Training Dynamics of the Cooldown Stage in Warmup-Stable-Decay Learning Rate Schedulercited by 12
  6. 2025Graph-Aware Isomorphic Attention for Adaptive Dynamics in Transformerscited by 7
  7. 2026Delta Attention Residualscited by 7
  8. 2026ACE-Merging: Data-Free Model Merging with Adaptive Covariance Estimationcited by 6
  9. 2025UniAttn: Reducing Inference Costs via Softmax Unification for Post-Training LLMscited by 3
  10. 2025Task Vectors, Learned Not Extracted: Performance Gains and Mechanistic Insightcited by 3
  11. 2024How Redundant Is the Transformer Stack in Speech Representation Models?cited by 2
  12. 2026Multi-Head Attention Residualscited by 2
11 more
  1. 2025Rethinking the Understanding Ability across LLMs through Mutual Informationcited by 1
  2. 2025Iterative Layer-wise Distillation for Efficient Compression of Large Language Modelscited by 1
  3. 2026Prometheus Mind: Retrofitting Memory to Frozen Language Modelscited by 1
  4. 2026The Scaling Properties of Implicit Deductive Reasoning in Transformerscited by 1
  5. 2025Beyond the Linear Separability Ceiling: Aligning Representations in VLMscited by 0
  6. 2025Activation Transport Operatorscited by 0
  7. 2026Variance Is Not Importance: Structural Analysis of Transformer Compressibility Across Model Scalescited by 0
  8. 2026Why Geometric Continuity Emerges in Deep Neural Networks: Residual Connections and Rotational Symmetry Breakingcited by 0
  9. 2026Dynamic Mode Decomposition along Depth in Vision Transformerscited by 0
  10. 2026Your Transformer Can Hold Two Thoughts at Once: Evidence of Linear Superposition in LLMscited by 0
  11. 2026The Residual Stream's Effective Depthcited by 0