Spectral Tail Interventions in Decoder-Only Language Models: Reasoning-Sensitive Weight Structure from Controlled Surgery
cs.LG
Submitted: 2026-08-09
Updated: 2026-08-09
Code: https://github.com/suzgunmirac/BIG-Bench-Hard
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Mistral 7B
- Scaling Laws for Neural Language Models
- From SGD to Spectra: A Theory of Neural Network Weight Dynamics
- In-context Learning and Induction Heads
- Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets
- Qwen2.5 Technical Report
- The Truth is in There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction
- Small Singular Values Matter: A Random Matrix Analysis of Transformer Models
- Spectral Surgery: Training-Free Refinement of LoRA via Gradient-Guided Singular Value Reweighting
- LLaMA: Open and Efficient Foundation Language Models
- Approaching Deep Learning through the Spectral Dynamics of Weights
- Towards Best Practices of Activation Patching in Language Models: Metrics and Methods
Related papers
- Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
- AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
- Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
- Convergence issues in Relational Concept Analysis based on AOC-posets
- Beliefs Beyond Posteriors: Local-Consistency Optimisation for Bayesian Neural Networks
- Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks