Block-Level Weight-Space Structure Persists Under Post-Training: An Empirical Study Across LLM Families
cs.LG
Submitted: 2026-08-23
Updated: 2026-08-23
Comments: 13 pages, 11 figures. Accepted at the ICML 2026 Workshop on Weight-Space Symmetries: from Foundations to Practical Applications. OpenReview: https://openreview.net/forum?id=jdVfddhT65
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- The Llama 3 Herd of Models
- Generalized Linear Mode Connectivity for Transformers
- Qwen2 Technical Report
- SGLang: Efficient Execution of Structured Language Model Programs
- Watch the Weights: Unsupervised monitoring and control of fine-tuned LLMs
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