Model Card for midi-model

Midi event transformer for music generation

Model Updates

  • tv2o-medium model: MidiTokenizerV2. 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

Training Details

Training Data

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
  • onnx/*.onnx: onnx format model
Downloads last month
248
Safetensors
Model size
233M params
Tensor type
F32
ยท
Inference API
Unable to determine this modelโ€™s pipeline type. Check the docs .

Model tree for skytnt/midi-model

Finetunes
1 model
Quantizations
2 models

Dataset used to train skytnt/midi-model

Spaces using skytnt/midi-model 16