README.md

---
title: Codette LoRA Fine-Tuning
license: mit
language:
- en
---

# Codette LoRA Fine-Tuning

Fine-tuning repo for Codette — a sovereign AI music production assistant built by Jonathan Harrison (Raiff's Bits).

## What This Does

Trains a LoRA adapter on top of `meta-llama/Llama-3.2-1B-Instruct` using Codette's own framework data, so she responds with her real voice, identity, and perspectives rather than as a generic assistant.

## Files

| File | Purpose |
|------|---------|
| `train_codette_lora.py` | Training script — runs as a HF Job |
| `codette_combined_train.jsonl` | 2,136 training examples from Codette's framework |

## Output

When training completes, the adapter is automatically pushed to:
**`Raiff1982/codette-llama-adapter`**

That adapter is then loaded by the Codette Space at:
**`Raiff1982/codette-ai`**

## Training Details

- **Base model**: meta-llama/Llama-3.2-1B-Instruct
- **Method**: LoRA (r=16, alpha=16)
- **Target modules**: q_proj, v_proj
- **Examples**: 2,136
- **Epochs**: 3
- **Hardware**: CPU (HF Jobs cpu-basic)

## Running the Job

See the HF Jobs documentation or follow the instructions in the Space README.