Transformer MLP Gate Thresholds Are Couplings to a Carried Reference Direction

arXiv:2609.31956 · cs.CL, cs.LG · Submitted 2026-09-25 · Read on arXiv

cs.CL, cs.LG

Submitted: 2026-09-25

Updated: 2026-09-25

Code: https://github.com/EvidentSolutions/llm-interp

Terminology

• Transformer — 16× in this paper · explained in 2 episodes
• Activation Sparsity — 5× in this paper · explained in 1 episode
• Sparsity — 5× in this paper · explained in 4 episodes
• Alignment — 3× in this paper · explained in 1 episode
• Adam — 2× in this paper · explained in 2 episodes
• Aggregation — 2× in this paper · explained in 1 episode
• Cosine Similarity — 2× in this paper · explained in 3 episodes
• Ablation — 1× in this paper · explained in 1 episode
• Decoupling — 1× in this paper · explained in 2 episodes
• Degeneracy — 1× in this paper · explained in 3 episodes
• In-Context Learning — 1× in this paper · explained in 4 episodes
• Linear probe — 1× in this paper · explained in 2 episodes
• Quantization — 1× in this paper · explained in 12 episodes

Sources

• The Discrete Charm of the MLP: Binary Routing of Continuous Signals in Transformer Feed-Forward Layers
• SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model
• The Curse and Blessing of Mean Bias in FP4-Quantized LLM Training
• Massive Spikes in LLMs are Bias Vectors: Mechanistic Uncovering and Spike-Free Quantization
• PaLM: Scaling Language Modeling with Pathways
• Hidden Dynamics of Massive Activations in Transformer Training
• Understanding Gated Neurons in Transformers from Their Input-Output Functionality
• Transient Reserves, Sink Dampers, and the Failure of Eigenvalue Reasoning in the Attention Propagator
• The Lazy Neuron Phenomenon: On Emergence of Activation Sparsity in Transformers
• Not a nuisance but a useful heuristic: Outlier dimensions favor frequent tokens in language models
• Steered LLM Activations are Non-Surjective
• A Refined Analysis of Massive Activations in LLMs
• A Unified View of Attention and Residual Sinks: Outlier-Driven Rescaling is Essential for Transformer Training
• Attention Sinks and Compression Valleys in LLMs are Two Sides of the Same Coin
• Qwen2.5 Technical Report
• A Single Layer to Explain Them All:Understanding Massive Activations in Large Language Models
• Bug or Feature$^2$: Weight Drift, Activation Sparsity and Spikes
• On the Relationship Between Activation Outliers and Feature Death in Sparse Autoencoders
• Massive Activations in Large Language Models
• The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks

Related papers

• Exploring Solution Divergence and Its Effect on Large Language Model Problem Solving
• Ishigaki-IDS-Bench: A Benchmark for Generating Information Delivery Specification from BIM Information Requirements
• Subliminal Steering: Stronger Encoding of Hidden Signals
• MedStruct-S: A Benchmark for Key Discovery, Key-Conditioned QA and Semi-Structured Extraction from OCR Clinical Reports
• The End of Transformers? On Challenging Attention and the Rise of Sub-Quadratic Architectures
• Untangling the Mechanisms of Misleading Context in Medical Question Answering