| --- |
| 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. |
