juancopi81 commited on
Commit
bfe9086
1 Parent(s): f2c092d

Training in progress epoch 0

Browse files
Files changed (3) hide show
  1. README.md +19 -93
  2. config.json +1 -1
  3. tf_model.h5 +1 -1
README.md CHANGED
@@ -1,126 +1,52 @@
1
  ---
2
  tags:
3
  - generated_from_keras_callback
4
- - music
5
  model-index:
6
  - name: juancopi81/mutopia_guitar_mmm
7
  results: []
8
- datasets:
9
- - juancopi81/mutopia_guitar_dataset
10
- widget:
11
- - text: "PIECE_START TIME_SIGNATURE=4_4 BPM=90 TRACK_START INST=0 DENSITY=2 BAR_START NOTE_ON=43"
12
- example_title: "Time signature 4/4, BPM=90, NOTE=G2"
13
  ---
14
 
15
- # juancopi81/mutopia_guitar_mmm
16
-
17
- Music generation could be approached similarly to language generation. There are many ways to represent music as text and then use a language model to create a model capable of music generation. For encoding MIDI files as text, I am using the excellent [implementation](https://github.com/AI-Guru/MMM-JSB) of Dr. Tristan Beheren of the paper: [MMM: Exploring Conditional Multi-Track Music Generation with the Transformer](https://arxiv.org/abs/2008.06048).
18
 
19
- This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on the [Mutopia Guitar Dataset](https://huggingface.co/datasets/juancopi81/mutopia_guitar_dataset). Use the widget to generate your piece, and then use [this notebook](https://colab.research.google.com/drive/14vlJwCvDmNH6SFfVuYY0Y18qTbaHEJCY?usp=sharing) to listen to the results (work in progress).
20
- I created the notebook as an adaptation of [the one created by Dr. Tristan Behrens](https://huggingface.co/TristanBehrens/js-fakes-4bars).
21
 
 
22
  It achieves the following results on the evaluation set:
23
- - Train Loss: 0.5365
24
- - Validation Loss: 1.5482
 
25
 
26
  ## Model description
27
 
28
- The model is GPT-2 loaded with the GPT2LMHeadModel architecture from Hugging Face. The context size is 256, and the vocabulary size is 588. The model uses a
29
- `WhitespaceSplit` pre-tokenizer. The [tokenizer](https://huggingface.co/juancopi81/mutopia_guitar_dataset_tokenizer) is also in the Hugging Face hub.
30
 
31
  ## Intended uses & limitations
32
 
33
- I built this model to learn more about how to use Hugging Face. I am implementing some of the parts of the [Hugging Face course](https://huggingface.co/course/chapter1/1) with a project that I find interesting.
34
- The main intention of this model is educational. I am creating a [series of notebooks](https://github.com/juancopi81/MMM_Mutopia_Guitar) where I show every step of the process:
35
- - Collecting the data
36
- - Pre-processing the data
37
- - Training a tokenizer from scratch
38
- - Fine-tuning a GPT-2 model
39
- - Building a Gradio app for the model
40
-
41
- I trained the model using the free version of Colab with a small dataset. Right now, it is heavily overfitting. My idea is to have a more extensive dataset of Guitar Music from Latinoamerica to train a new model similar to the Mutopia Guitar Model, using more GPU resources.
42
 
43
  ## Training and evaluation data
44
 
45
- I am training the model with [Mutopia Guitar Dataset](https://huggingface.co/datasets/juancopi81/mutopia_guitar_dataset). This dataset consists of the soloist guitar pieces of the [Mutopia Project](https://www.mutopiaproject.org/).
46
- The dataset mainly contains guitar music from western classical composers, such as Sor, Aguado, Carcassi, and Giuliani.
47
 
48
- For the first epochs of training, I transposed the notes by raising and lowering the pitches using the twelve semi-tones of an entire octave. Later, I trained the model without transposing the pieces so that generation shows better results of a real guitar piece.
49
 
50
  ### Training hyperparameters
51
 
52
- The following hyperparameters were used during training (with transposition):
53
- - optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'WarmUp', 'config': {'initial_learning_rate': 5e-07, 'decay_schedule_fn': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-07, 'decay_steps': 5726, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, 'passive_serialization': True}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
54
-
55
- The following hyperparameters were used during training (without transposition - first round):
56
  - optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'WarmUp', 'config': {'initial_learning_rate': 5e-07, 'decay_schedule_fn': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-07, 'decay_steps': 350, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, '__passive_serialization__': True}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
57
-
58
- The following hyperparameters were used during training (without transposition - second round):
59
- - optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'WarmUp', 'config': {'initial_learning_rate': 5e-07, 'decay_schedule_fn': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-07, 'decay_steps': 350, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, '__passive_serialization__': True}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
60
-
61
- The following hyperparameters were used during training (without transposition, new tokenizer - third round):
62
- - optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'WarmUp', 'config': {'initial_learning_rate': 5e-07, 'decay_schedule_fn': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-07, 'decay_steps': 350, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, 'passive_serialization': True}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
63
-
64
  - training_precision: mixed_float16
 
65
  ### Training results
66
- Using transposition:
67
- | Train Loss | Validation Loss | Epoch |
68
- |:----------:|:---------------:|:-----:|
69
- | 1.0705 | 1.3590 | 0 |
70
- | 0.8889 | 1.3702 | 1 |
71
- | 0.7588 | 1.3974 | 2 |
72
- | 0.7294 | 1.4813 | 3 |
73
- | 0.6263 | 1.5263 | 4 |
74
- | 0.5841 | 1.5263 | 5 |
75
- | 0.5844 | 1.5263 | 6 |
76
- | 0.5837 | 1.5346 | 7 |
77
- | 0.5798 | 1.5411 | 8 |
78
- | 0.5773 | 1.5440 | 9 |
79
-
80
- Without transposition (first round):
81
- | Train Loss | Validation Loss | Epoch |
82
- |:----------:|:---------------:|:-----:|
83
- | 0.5503 | 1.5436 | 0 |
84
- | 0.5503 | 1.5425 | 1 |
85
- | 0.5476 | 1.5425 | 2 |
86
- | 0.5467 | 1.5425 | 3 |
87
- | 0.5447 | 1.5431 | 4 |
88
- | 0.5418 | 1.5447 | 5 |
89
- | 0.5418 | 1.5451 | 6 |
90
- | 0.5401 | 1.5472 | 7 |
91
- | 0.5386 | 1.5479 | 8 |
92
- | 0.5365 | 1.5482 | 9 |
93
-
94
- Without transposition (second round):
95
- | Train Loss | Validation Loss | Epoch |
96
- |:----------:|:---------------:|:-----:|
97
- | 0.5368 | 1.5482 | 0 |
98
- | 0.5355 | 1.5480 | 1 |
99
- | 0.5326 | 1.5488 | 2 |
100
- | 0.5363 | 1.5493 | 3 |
101
- | 0.5346 | 1.5488 | 4 |
102
- | 0.5329 | 1.5502 | 5 |
103
- | 0.5329 | 1.5514 | 6 |
104
- | 0.5308 | 1.5514 | 7 |
105
- | 0.5292 | 1.5536 | 8 |
106
- | 0.5272 | 1.5543 | 9 |
107
-
108
- Without transposition (third round - new tokenizer):
109
  | Train Loss | Validation Loss | Epoch |
110
  |:----------:|:---------------:|:-----:|
111
- | 6.1361 | 6.4569 | 0 |
112
- | 5.6383 | 5.8249 | 1 |
113
- | 4.9125 | 4.8956 | 2 |
114
- | 4.2013 | 4.2778 | 3 |
115
- | 3.8665 | 4.0330 | 4 |
116
- | 3.7106 | 3.8956 | 5 |
117
- | 3.6041 | 3.7995 | 6 |
118
- | 3.5301 | 3.7485 | 7 |
119
- | 3.4973 | 3.7323 | 8 |
120
- | 3.4909 | 3.7323 | 9 |
121
 
122
  ### Framework versions
123
- - Transformers 4.22.1
 
124
  - TensorFlow 2.8.2
125
  - Datasets 2.5.1
126
- - Tokenizers 0.12.1
 
1
  ---
2
  tags:
3
  - generated_from_keras_callback
 
4
  model-index:
5
  - name: juancopi81/mutopia_guitar_mmm
6
  results: []
 
 
 
 
 
7
  ---
8
 
9
+ <!-- This model card has been generated automatically according to the information Keras had access to. You should
10
+ probably proofread and complete it, then remove this comment. -->
 
11
 
12
+ # juancopi81/mutopia_guitar_mmm
 
13
 
14
+ This model was trained from scratch on an unknown dataset.
15
  It achieves the following results on the evaluation set:
16
+ - Train Loss: 3.4879
17
+ - Validation Loss: 3.7206
18
+ - Epoch: 0
19
 
20
  ## Model description
21
 
22
+ More information needed
 
23
 
24
  ## Intended uses & limitations
25
 
26
+ More information needed
 
 
 
 
 
 
 
 
27
 
28
  ## Training and evaluation data
29
 
30
+ More information needed
 
31
 
32
+ ## Training procedure
33
 
34
  ### Training hyperparameters
35
 
36
+ The following hyperparameters were used during training:
 
 
 
37
  - optimizer: {'name': 'AdamWeightDecay', 'learning_rate': {'class_name': 'WarmUp', 'config': {'initial_learning_rate': 5e-07, 'decay_schedule_fn': {'class_name': 'PolynomialDecay', 'config': {'initial_learning_rate': 5e-07, 'decay_steps': 350, 'end_learning_rate': 0.0, 'power': 1.0, 'cycle': False, 'name': None}, '__passive_serialization__': True}, 'warmup_steps': 1000, 'power': 1.0, 'name': None}}, 'decay': 0.0, 'beta_1': 0.9, 'beta_2': 0.999, 'epsilon': 1e-08, 'amsgrad': False, 'weight_decay_rate': 0.01}
 
 
 
 
 
 
 
38
  - training_precision: mixed_float16
39
+
40
  ### Training results
41
+
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
42
  | Train Loss | Validation Loss | Epoch |
43
  |:----------:|:---------------:|:-----:|
44
+ | 3.4879 | 3.7206 | 0 |
45
+
 
 
 
 
 
 
 
 
46
 
47
  ### Framework versions
48
+
49
+ - Transformers 4.22.2
50
  - TensorFlow 2.8.2
51
  - Datasets 2.5.1
52
+ - Tokenizers 0.12.1
config.json CHANGED
@@ -32,7 +32,7 @@
32
  "max_length": 350
33
  }
34
  },
35
- "transformers_version": "4.22.1",
36
  "use_cache": true,
37
  "vocab_size": 588
38
  }
 
32
  "max_length": 350
33
  }
34
  },
35
+ "transformers_version": "4.22.2",
36
  "use_cache": true,
37
  "vocab_size": 588
38
  }
tf_model.h5 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c763c4bb92abe5f89ca5aa833c7476c850feb7673165f1c795569eeefaec6f3f
3
  size 345352296
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:65a22e285ce87a9ea0b11fa05ff56f4ef6faa84566fd03de23b9b7f6ea1003b2
3
  size 345352296