Are we Merging the Right Models? Impact of Expert Training Duration on Model Merging for LLMs
Nikita Kozodoi, Zainab Afolabi, Jack Butler
cs.LG, cs.AI, stat.ML
Submitted: 2026-07-13
Comments: Accepted to ICML 2026 workshop on weight-space symmetries
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone
- Tulu 3: Pushing Frontiers in Open Language Model Post-Training
- Small Language Models are the Future of Agentic AI
- Magicoder: Empowering Code Generation with OSS-Instruct
- HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusal
- Evaluating Large Language Models Trained on Code
- Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge
- Training Verifiers to Solve Math Word Problems
- Post-Hoc Reversal: Are We Selecting Models Prematurely?
- (How) Learning Rates Regulate Catastrophic Overtraining
- MergeBench: A Benchmark for Merging Domain-Specialized LLMs
- Aya Dataset: An Open-Access Collection for Multilingual Instruction Tuning
- From Memorization to Parameter Interference: How Overtraining Experts Harms Model Merging
- Leveraging Model Soups to Classify Intangible Cultural Heritage Images from the Mekong Delta
- DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving
- Instruction-Following Evaluation for Large Language Models
- ATM: Improving Model Merging by Alternating Tuning and Merging
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