--- tags: - generated_from_keras_callback - music model-index: - name: juancopi81/mutopia_guitar_mmm results: [] datasets: - juancopi81/mutopia_guitar_dataset widget: - text: "PIECE_START TIME_SIGNATURE=4_4 BPM=90 TRACK_START INST=0 DENSITY=2 BAR_START NOTE_ON=43" example_title: "Time signature 4/4, BPM=90, NOTE=G2" --- # juancopi81/mutopia_guitar_mmm 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). 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). I created the notebook as an adaptation of [the one created by Dr. Tristan Behrens](https://huggingface.co/TristanBehrens/js-fakes-4bars). It achieves the following results on the evaluation set: - Train Loss: 0.5365 - Validation Loss: 1.5482 ## Model description 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 `WhitespaceSplit` pre-tokenizer. The [tokenizer](https://huggingface.co/juancopi81/mutopia_guitar_dataset_tokenizer) is also in the Hugging Face hub. ## Intended uses & limitations 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. 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: - Collecting the data - Pre-processing the data - Training a tokenizer from scratch - Fine-tuning a GPT-2 model - Building a Gradio app for the model 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. ## Training and evaluation data 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/). The dataset mainly contains guitar music from western classical composers, such as Sor, Aguado, Carcassi, and Giuliani. 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. ### Training hyperparameters The following hyperparameters were used during training (with transposition): - 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} The following hyperparameters were used during training (without transposition - first round): - 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} The following hyperparameters were used during training (without transposition - second round): - 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} - training_precision: mixed_float16 ### Training results Using transposition: | Train Loss | Validation Loss | Epoch | |:----------:|:---------------:|:-----:| | 1.0705 | 1.3590 | 0 | | 0.8889 | 1.3702 | 1 | | 0.7588 | 1.3974 | 2 | | 0.7294 | 1.4813 | 3 | | 0.6263 | 1.5263 | 4 | | 0.5841 | 1.5263 | 5 | | 0.5844 | 1.5263 | 6 | | 0.5837 | 1.5346 | 7 | | 0.5798 | 1.5411 | 8 | | 0.5773 | 1.5440 | 9 | Without transposition (first round): | Train Loss | Validation Loss | Epoch | |:----------:|:---------------:|:-----:| | 0.5503 | 1.5436 | 0 | | 0.5503 | 1.5425 | 1 | | 0.5476 | 1.5425 | 2 | | 0.5467 | 1.5425 | 3 | | 0.5447 | 1.5431 | 4 | | 0.5418 | 1.5447 | 5 | | 0.5418 | 1.5451 | 6 | | 0.5401 | 1.5472 | 7 | | 0.5386 | 1.5479 | 8 | | 0.5365 | 1.5482 | 9 | Without transposition (second round): | Train Loss | Validation Loss | Epoch | |:----------:|:---------------:|:-----:| | 0.5368 | 1.5482 | 0 | | 0.5355 | 1.5480 | 1 | | 0.5326 | 1.5488 | 2 | | 0.5363 | 1.5493 | 3 | | 0.5346 | 1.5488 | 4 | | 0.5329 | 1.5502 | 5 | | 0.5329 | 1.5514 | 6 | | 0.5308 | 1.5514 | 7 | | 0.5292 | 1.5536 | 8 | | 0.5272 | 1.5543 | 9 | ### Framework versions - Transformers 4.22.1 - TensorFlow 2.8.2 - Datasets 2.5.1 - Tokenizers 0.12.1