https://github.com/willibrandon/lofi-lora
https://github.com/willibrandon/lofi-lora
Last synced: 13 days ago
JSON representation
- Host: GitHub
- URL: https://github.com/willibrandon/lofi-lora
- Owner: willibrandon
- Created: 2025-12-23T01:25:16.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2025-12-23T05:26:22.000Z (5 months ago)
- Last Synced: 2025-12-24T15:26:27.446Z (5 months ago)
- Language: Shell
- Size: 97.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# lofi-lora
Custom ACE-Step LoRA training for lofi music generation.
## Overview
This repository contains the training pipeline for a lofi-optimized LoRA adapter for [ACE-Step](https://github.com/ace-step/ACE-Step). The trained model enhances lofi beat generation with authentic vinyl warmth, jazz influences, and genre-specific characteristics across 30+ style variations.
## Related Projects
- [lofi.nvim](https://github.com/willibrandon/lofi.nvim) - Neovim plugin for AI-generated lofi beats
- [willibrandon/lofi-models](https://huggingface.co/willibrandon/lofi-models) - Trained model weights
- [ACE-Step](https://github.com/ace-step/ACE-Step) - Base model and training framework
## Documentation
See [docs/lofi-lora-training.md](docs/lofi-lora-training.md) for the full training specification including:
- Data curation strategy and legal sources
- Style taxonomy (30+ lofi variations)
- Prompt engineering system
- Training configuration
- Model optimization (quantization)
- Quality validation
## Quick Start
```bash
# Clone repos
git clone https://github.com/willibrandon/lofi-lora.git
git clone https://github.com/ace-step/ACE-Step.git
# Install ACE-Step
cd ACE-Step && pip install -e . && cd ..
# Install dependencies
cd lofi-lora && pip install -r requirements.txt
# Prepare data (see docs for data acquisition)
# Place tracks in data/ with corresponding _prompt.txt and _lyrics.txt files
# Convert to HuggingFace dataset
python scripts/convert_dataset.py --data_dir ./data --output_name lofi_lora_dataset
# Train
./train.sh
```
## Hardware Requirements
- GPU: RTX 4090 (24GB VRAM) or equivalent
- RAM: 64GB+
- Storage: 2TB+ NVMe
- Training time: ~60 hours for 100k steps
## License
Apache 2.0