emerging-techniques-model-merging — Claude Code skill
A Claude skill that walks through merging fine-tuned models with mergekit, so you can blend capabilities without retraining.
npx claude-code-templates@latest --skill ai-research/emerging-techniques-model-mergingWhat emerging-techniques-model-merging does
This skill covers combining several fine-tuned models into one using mergekit. The usual case: you have separate models for math, code and chat, and want a single one that handles all three. No retraining is involved, since the weights are simply combined.
It goes through the merge methods: linear averaging, SLERP for a pair of models, Task Arithmetic built on task vectors, TIES-Merging for sign conflicts, and DARE for dropping redundant deltas. Each comes with a YAML config example and its parameters, such as weight, density and t.
You need mergekit, transformers and torch installed. mergekit can be cloned from arcee-ai or pulled with pip install mergekit. Merges run on CPU, so a GPU is not required and trying out variants stays cheap.
What it helps with
- blend a math model, a coding model and a chat model into one
- produce a new model variant without GPU training
- resolve weight sign conflicts when merging several models
- drop redundant deltas to cut down on noise
- test a few merge configs in a row
Example prompts for the agent
- Write a mergekit config for a linear merge of two models with equal weights
- Set up a SLERP merge of Mistral and OpenHermes with t at 0.5
- Configure TIES-Merging for three models with density 0.5
How to install emerging-techniques-model-merging
1claude-code-templates
npx claude-code-templates@latest --skill ai-research/emerging-techniques-model-merging2Skills CLI (Vercel)
npx skills add davila7/claude-code-templates --skill emerging-techniques-model-merging -a claude-code3Manually: put the files in place
~/.claude/skills/emerging-techniques-model-merging/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/emerging-techniques-model-merging/
Manually: put the files in place
.agents/skills/emerging-techniques-model-merging/SKILL.mdFor all projects use ~/.agents/skills/
Manually: put the files in place
.cursor/skills/emerging-techniques-model-merging/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install davila7/claude-code-templates emerging-techniques-model-merging2Manually: put the files in place
.github/skills/emerging-techniques-model-merging/SKILL.mdPlatform CLI
gemini skills install https://github.com/davila7/claude-code-templates --path cli-tool/components/skills/ai-research/emerging-techniques-model-mergingManually: put the files in place
.devin/skills/emerging-techniques-model-merging/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about emerging-techniques-model-merging
How do I install emerging-techniques-model-merging in Claude Code?
Run npx claude-code-templates@latest --skill ai-research/emerging-techniques-model-merging. Alternatively use npx skills add davila7/claude-code-templates --skill emerging-techniques-model-merging -a claude-code. The skill file lands at ~/.claude/skills/emerging-techniques-model-merging/SKILL.md.
What does emerging-techniques-model-merging do?
It helps you merge fine-tuned models with mergekit to combine their abilities without retraining. It covers linear, SLERP, Task Arithmetic, TIES-Merging and DARE, each with config examples.
What do I need to run it?
The mergekit, transformers and torch packages. Merges execute on CPU, so no GPU is needed.