Unsloth Buddy
Fine-tune and evaluate open models with Unsloth and MLX helpers scripted for Claude-driven ML experiments on Mac or Colab.
Overview
Unsloth-buddy is a plugin marketplace for the Build phase that connects Claude Code to Unsloth and MLX fine-tuning scripts, eval templates, and demo tooling for custom LLMs.
What is this marketplace?
- Single unsloth-buddy plugin with bundled SKILL.md, AGENTS.md, and many training scripts.
- Example flows for SFT, DPO, GRPO, vision, and MLX variants including Apple Silicon paths.
- Init and detect scripts (init_project, detect_system, detect_env) for environment-aware setup.
- Dashboard and demo templates plus colab_training for notebook-first indies.
- Gaslamp-branded callbacks and reflect helpers for experiment traceability.
- 1 plugin in marketplace
- 20+ bundled scripts listed in marketplace manifest
- Publisher: Gaslamp (https://gaslamp.dev)
Community signal: 249 GitHub stars.
What problem does it solve?
Indie builders want custom models but get stuck wiring Unsloth, MLX, evals, and demos across fragmented scripts and notebooks.
Who is it for?
Solo ML-curious builders fine-tuning open models on Mac MLX or Colab with Claude Code orchestrating scripts.
Skip if: Teams needing enterprise MLOps, managed GPU clusters, or production model serving SLAs without hands-on training work.
What do I get? / Deliverables
You get an agent-callable project scaffold, training recipes, and dashboards so fine-tune experiments move from idea to measurable demo faster.
- Initialized fine-tuning project via init_project.py
- Runnable SFT/DPO/GRPO or vision training runs from bundled examples
- Demo server or dashboard HTML for qualitative eval
Plugins in this marketplace
1 plugin — install individually after you add the marketplace.
Recommended Marketplaces
Journey fit
Canonical shelf is build/agent-tooling because the marketplace wires agent workflows to training, eval, and demo scripts rather than production hosting alone. Agent-tooling captures SKILL.md-driven harnesses, callback dashboards, and example pipelines the coding agent invokes directly.
How it compares
Fine-tuning agent harness with scripts, not a hosted training API or generic prompt-only skill.
Common Questions / FAQ
Who is Unsloth Buddy for?
Builders shipping agent or API products who want Claude to drive Unsloth fine-tuning, eval, and local demos.
When should I use Unsloth Buddy?
Use it when validating whether a fine-tuned open model solves your niche better than frontier APIs and you are ready to run training scripts.
How do I add Unsloth Buddy to my agent?
Install the Unsloth Buddy marketplace plugin in Claude Code, then follow SKILL.md and AGENTS.md with the bundled init and detect scripts.