Get Started
Home
Topics
Search
Library
Research questionHow can multiple task-specific LLMs be merged without full retraining or parameter interference?Combining separately adapted models can make their parameter updates conflict, causing the merged model to lose capabilities that the individual models had. The difficulty is to form one multi-task model without incurring the cost of retraining all parameters.
AI
LLM Pretraining & Post-training
Machine Learning
Natural Language Processing
Latest papersRecent research connected to this question, newest first.CoMerge: Conflict-Driven Preference Optimization for Multi-Task Model MergingThe source concerns model merging for large language models and reports evidence on MergeBench and Llama-3.1-8B-Instruct, including instruction-following and safety performance. Its results compare against data-free and data-driven merging baselines and use lightweight coefficient optimization rather than full-parameter retraining.research paper · Sep 2, 2026
Related questions
How can LLM agents generalize to unseen tasks without directly fine-tuning their policies?How can LLM agents jointly adapt reasoning policies and hierarchical skill libraries during reinforcement learning?How can one deployable LLM learn from multiple specialized teachers when the most reliable teacher varies by sample?How can LLM pretraining avoid sudden gradient explosions when scaling to larger models?