---
library_name: mlx
pipeline_tag: text-generation
tags:
- music
- midi
- generation
- mlx
- autoregressive
- structure
- musiclang
---

# okai-musiclang-structure v2.0 - Structure Model

This is an autoregressive structure generation model for music using MLX.

## Model Details

- **Model Type**: Structure Generator (Autoregressive)
- **Version**: v2.0
- **Step**: 3000
- **Architecture**: Transformer with causal language modeling
- **Vocabulary Size**: 4796
- **Model Dimension**: 256
- **Layers**: 6
- **Max Sequence Length**: 1024

## Training Configuration

- **Batch Size**: 16
- **Learning Rate**: 1e-05
- **Training Steps**: 3000

## Usage

This model generates sequential music structure:
- Input: Song control tokens (genre, instruments, etc.)
- Output: Sequential bar structure with chords and tonality

Example generation:
```
GENRE__ROCK SUBGENRE__ALTERNATIVE START BAR__1 CHORD_DEGREE__1 TONALITY_DEGREE__1 BAR__2 CHORD_DEGREE__4 TONALITY_DEGREE__5 ... WILL_END
```

Generated with MLX framework for Apple Silicon.