Weight-Adjusted Gradients Reveal Parameter Importance and Failure Modes in LLMs

arXiv:2607.10803 · cs.LG, cs.AI · Submitted 2026-07-12 · Read on arXiv

cs.LG, cs.AI

Submitted: 2026-07-12

Updated: 2026-09-29

Terminology

• Weight-Adjusted Gradients — 2× in this paper
• Failure Modes — 1× in this paper · explained in 3 episodes
• Interpretability — 1× in this paper · explained in 15 episodes
• Large Language Models (LLMs) — 1× in this paper · explained in 6 episodes
• Quantization — 1× in this paper · explained in 12 episodes

Sources

• Developing and Maintaining an Open-Source Repository of AI Evaluations: Challenges and Insights
• Layer Normalization
• Exposing the Illusion of Erasure in Knowledge Editing for LLMs
• Evaluating Large Language Models Trained on Code
• Training Verifiers to Solve Math Word Problems
• Beyond Size: How Gradients Shape Pruning Decisions in Large Language Models
• Golden Layers and Where to Find Them: Improved Knowledge Editing for Large Language Models Via Layer Gradient Analysis
• A Primer on the Inner Workings of Transformer-based Language Models
• The State of Sparsity in Deep Neural Networks
• Scaling Laws for Neural Language Models
• Large Language Models: A Survey
• GPT-4 Technical Report
• How Many Parameters Does Your Task Really Need? Task Specific Pruning with LLM-Sieve
• Understanding Performance Collapse in Layer-Pruned Large Language Models via Decision Representation Transitions
• The Curse of Recursion: Training on Generated Data Makes Models Forget
• Gemma 3 Technical Report
• LLaMA: Open and Efficient Foundation Language Models
• Qwen3 Technical Report
• Boosting Large Language Models with Mask Fine-Tuning
• A Comprehensive Study of Knowledge Editing for Large Language Models

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: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks