Model Merging

If you want to combine multiple fine-tuned models without spinning up GPUs or retraining from scratch, this gets you there. It's built around mergekit and lets you blend domain expertise from different models (say, a math specialist with a coding specialist) to create something new in minutes instead of days. The approach has real track record: Marcoro14-7B-slerp topped the Open LLM Leaderboard in early 2024 using these techniques. Expect 5-10% benchmark improvements in many cases, and you avoid the catastrophic forgetting problem that kills multi-task fine-tuning. Runs entirely on CPU, which means you can experiment with model variants without burning cloud credits on compute.

Install to Claude Code

npx -y skills add davila7/claude-code-templates --skill model-merging --agent claude-code