The Smallest Singular Directions of the Unembedding Matter Disproportionately for Next-Token Prediction
cs.LG, cs.CL
Submitted: 2026-05-16
Updated: 2026-11-03
Terminology
Sources
- Seq-VCR: Preventing Collapse in Intermediate Transformer Representations for Enhanced Reasoning
- Eliciting Latent Predictions from Transformers with the Tuned Lens
- Gemma 2: Improving Open Language Models at a Practical Size
- Disentangling Geometry, Performance, and Training in Language Models
- Tracing the Representation Geometry of Language Models from Pretraining to Post-training
- Textbooks Are All You Need II: phi-1.5 technical report
- The Geometry of Categorical and Hierarchical Concepts in Large Language Models
- Why Linear Interpretability Works: Invariant Subspaces as a Result of Architectural Constraints
- Opening the Black Box of Deep Neural Networks via Information
- Layer by Layer: Uncovering Hidden Representations in Language Models
- Implicit Geometry of Next-token Prediction: From Language Sparsity Patterns to Model Representations
Related papers
- A Physics-Regulated Neural Framework for Learning 3D Grain Growth Dynamics
- FlexMoE: Learning One-for-All Multi-Budget Expert Pruning for Flexible MoE Deployment
- OptiMind: Improving Small Language Models for Optimization Modeling through Formulation Guidance and Multi-Turn Learning
- Staying in the Sweet Spot: Boost Reasoning Evolution via Responsive Rollout Difficulty Manipulation
- Examining the Impact of Data and Training Methods on Emergent and Subliminal Misalignment
- Rethinking Black-Box On-Policy Distillation: A Rubric-based Analysis