>OPEN SOURCE MODEL TUNING

Nanotune

A simple, interactive CLI for fine-tuning small language models on Apple Silicon. No YAML configs, no complex flags - just an interactive TUI.

Nanotune - Training
Model: Qwen/Qwen2.5-Coder-1.5B-Instruct
Examples: 782
Iterations: 150
Progress:████████████████████100%
Iteration: 150/150
Training Loss
1.070.15
Train Loss: 0.1500
[Ctrl+C] Stop training (checkpoint saved)

Not for profit. For the community.

Nanotune is a simple, interactive CLI for fine-tuning small language models on Apple Silicon. No YAML configs, no complex flags - just an interactive TUI that guides you through each step.

Built by the Nano Collective, a community collective building AI tooling not for profit, but for the community. Everything we build is open, transparent, and driven by the people who use it.

Premium capabilities.
Zero paid tiers.

Apple Silicon Native

Optimized for MLX to leverage unified memory architecture, delivering blazing fast local fine-tuning on Mac.

Interactive TUI

No complex YAML configs or endless CLI flags. A clean, interactive interface guides you through every step.

End-to-End Pipeline

Automatically handles dataset validation, ChatML formatting, training loops, and LoRA fusion.

Built-In Evaluation

Validate model performance with integrated LLM judge and benchmarking suites directly from the CLI.

GGUF Export

Export your tuned models to standard GGUF format for direct use in Ollama, LM Studio, or llama.cpp.

Instant Chat REPL

Test your fine-tuned models immediately via the built-in chat interface. No external tools required.

Privacy First

Train models entirely offline. Your proprietary datasets never leave your machine.

Terminal Native

Designed explicitly for terminal workflows. Fits seamlessly into your existing developer environment.

Realistic workflows.

Nanotune handles dataset formatting, local fine-tuning, and inference testing natively in your terminal.

Dataset
$ nanotune data add
Nanotune - Add Training Data
Examples: 12 | Building: 1 turn
User input:
How do I reverse a list in Python?
Turn added!
[Enter] Submit [Tab] Switch [Esc] Save & exit
Train
$ nanotune train --model mistral-7b
Nanotune - Training
Progress: ███████████░░░░░ 72%
Iteration: 108/150
Train Loss: 0.184 · Val Loss: 0.206
[Ctrl+C] Stop training (checkpoint saved)
Chat
$ nanotune chat
Nanotune - Chat
Model: mistral-7b-q4_k_m.gguf • System: system
You:
What is a tensor?
Model:
A tensor is a multi-dimensional array that generalizes scalars, vectors, and matrices...
TTFT 187ms · 18.3 tok/s · 92 tokens

Install in seconds.

Zero complex setup. Available on your favorite package manager.

NPM
npm install -g @nanocollective/nanotune
NPX
npx @nanocollective/nanotune init

Why developers choose Nanotune

100% Open Source

Read the code. Audit the code. Improve the code. It belongs to the community, not a corporation.

Apple Silicon Optimized

Built specifically for M-series chips to maximize training efficiency using MLX.

Zero Vendor Lock-In

Export your models to universally supported formats. Your workflow remains entirely yours.

Absolute Privacy

Fine-tune on sensitive company data without fear of leaks or API terms of service changes.

Community Driven

Features are built based on what developers actually need, not what looks good in a marketing pitch.

Highly Extensible

Seamlessly integrates with the broader MLX and llama.cpp ecosystems for advanced use cases.

Supported by

Sponsorship funds the people building open, privacy-respecting AI tools, and keeps every line of it free at the point of use.

Become a sponsor
[ Join the Collective ]

Build with us.

Join developers building the future of local-first AI tooling.