Upload structure model checkpoint at step 12279
Browse files- checkpoint_12279/README.md +46 -0
- checkpoint_12279/config.json +14 -0
- checkpoint_12279/model.npz +3 -0
- checkpoint_12279/training_state.json +26 -0
checkpoint_12279/README.md
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
library_name: mlx
|
| 3 |
+
pipeline_tag: text-generation
|
| 4 |
+
tags:
|
| 5 |
+
- music
|
| 6 |
+
- midi
|
| 7 |
+
- generation
|
| 8 |
+
- mlx
|
| 9 |
+
- autoregressive
|
| 10 |
+
- structure
|
| 11 |
+
- musiclang
|
| 12 |
+
---
|
| 13 |
+
|
| 14 |
+
# okai-musiclang-structure v2.0 - Structure Model
|
| 15 |
+
|
| 16 |
+
This is an autoregressive structure generation model for music using MLX.
|
| 17 |
+
|
| 18 |
+
## Model Details
|
| 19 |
+
|
| 20 |
+
- **Model Type**: Structure Generator (Autoregressive)
|
| 21 |
+
- **Version**: v2.0
|
| 22 |
+
- **Step**: 12279
|
| 23 |
+
- **Architecture**: Transformer with causal language modeling
|
| 24 |
+
- **Vocabulary Size**: 4796
|
| 25 |
+
- **Model Dimension**: 256
|
| 26 |
+
- **Layers**: 6
|
| 27 |
+
- **Max Sequence Length**: 1024
|
| 28 |
+
|
| 29 |
+
## Training Configuration
|
| 30 |
+
|
| 31 |
+
- **Batch Size**: 8
|
| 32 |
+
- **Learning Rate**: 0.0001
|
| 33 |
+
- **Training Steps**: 12279
|
| 34 |
+
|
| 35 |
+
## Usage
|
| 36 |
+
|
| 37 |
+
This model generates sequential music structure:
|
| 38 |
+
- Input: Song control tokens (genre, instruments, etc.)
|
| 39 |
+
- Output: Sequential bar structure with chords and tonality
|
| 40 |
+
|
| 41 |
+
Example generation:
|
| 42 |
+
```
|
| 43 |
+
GENRE__ROCK SUBGENRE__ALTERNATIVE START BAR__1 CHORD_DEGREE__1 TONALITY_DEGREE__1 BAR__2 CHORD_DEGREE__4 TONALITY_DEGREE__5 ... WILL_END
|
| 44 |
+
```
|
| 45 |
+
|
| 46 |
+
Generated with MLX framework for Apple Silicon.
|
checkpoint_12279/config.json
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_name": "okai-musiclang-structure",
|
| 3 |
+
"model_version": "v2.0",
|
| 4 |
+
"model_type": "structure",
|
| 5 |
+
"global_step": 12279,
|
| 6 |
+
"architecture": "AutoregressiveTransformer",
|
| 7 |
+
"training_type": "causal_lm",
|
| 8 |
+
"vocab_size": 4796,
|
| 9 |
+
"model_dim": 256,
|
| 10 |
+
"num_heads": 8,
|
| 11 |
+
"num_layers": 6,
|
| 12 |
+
"max_sequence_length": 1024,
|
| 13 |
+
"dropout": 0.1
|
| 14 |
+
}
|
checkpoint_12279/model.npz
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a9b11ed83beedec1c67a337e4b22c16fd0200e5665f712eb4a9a1ab8cba7cafb
|
| 3 |
+
size 29839884
|
checkpoint_12279/training_state.json
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"global_step": 12279,
|
| 3 |
+
"final_epoch": 6,
|
| 4 |
+
"training_config": {
|
| 5 |
+
"batch_size": 8,
|
| 6 |
+
"num_epochs": 6,
|
| 7 |
+
"learning_rate": 0.0001,
|
| 8 |
+
"weight_decay": 0.01,
|
| 9 |
+
"warmup_steps": 1000,
|
| 10 |
+
"max_grad_norm": 1.0,
|
| 11 |
+
"eval_steps": 500,
|
| 12 |
+
"save_steps": 500
|
| 13 |
+
},
|
| 14 |
+
"model_config": {
|
| 15 |
+
"vocab_size": 4796,
|
| 16 |
+
"model_dim": 256,
|
| 17 |
+
"num_heads": 8,
|
| 18 |
+
"num_layers": 6,
|
| 19 |
+
"max_sequence_length": 1024,
|
| 20 |
+
"dropout": 0.1
|
| 21 |
+
},
|
| 22 |
+
"model_name": "okai-musiclang-structure",
|
| 23 |
+
"model_version": "v2.0",
|
| 24 |
+
"model_type": "structure",
|
| 25 |
+
"final_model": true
|
| 26 |
+
}
|