File size: 1,406 Bytes
54fe4e7
 
ae617e5
 
 
 
 
 
 
 
54fe4e7
ae617e5
 
 
 
09d9f65
615d8a8
180992f
 
615d8a8
ae617e5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
615d8a8
 
ae617e5
615d8a8
ae617e5
 
 
615d8a8
ae617e5
 
 
 
b5fb5d4
615d8a8
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
---
license: apache-2.0
datasets:
- projectlosangeles/Los-Angeles-MIDI-Dataset
library_name: transformers
tags:
- music
- midi
- music generation
- midi generation
---
# Model Card for midi-model

Midi event transformer for music generation

### Model Updates

- tv2o-medium model: MidiTokenizerV2. [skytnt/midi-model-tv2o-medium](https://huggingface.co./skytnt/midi-model-tv2o-medium)
- v1.2 : Optimise the tokenizer and dataset. The dataset was filtered by MIDITokenizer.check_quality. Using the higher quality dataset to train the model, the performance of the model is significantly improved.

## Model Details

### Model Description

- **Developed by:** SkyTNT
- **Model type:** Transformer
- **License:** apache-2.0

### Model Sources

- **Repository:** https://github.com/SkyTNT/midi-model
- **Demo:** https://huggingface.co./spaces/skytnt/midi-composer


## Training Details

### Training Data

- [projectlosangeles/Los-Angeles-MIDI-Dataset](https://huggingface.co./datasets/projectlosangeles/Los-Angeles-MIDI-Dataset)

#### Training Hyperparameters

- basemodel: v1.1
- lr: 1e-4
- weight-decay: 0.01
- batch: 2x2x2
- fp16 mixed precision

####  Loss
 - val loss: 0.238

## Files

- model.ckpt: latest model checkpoint
- soundfont.sf2: soundfont file, i put it here for convenient downloading , [source](https://github.com/musescore/MuseScore/tree/master/share/sound)
- onnx/*.onnx: onnx format model