Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,3 +1,26 @@
|
|
| 1 |
---
|
| 2 |
license: mit
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: mit
|
| 3 |
+
tags:
|
| 4 |
+
- symbolic-music
|
| 5 |
+
- music-infilling
|
| 6 |
+
- midi
|
| 7 |
+
- masked-language-model
|
| 8 |
+
library_name: pytorch
|
| 9 |
---
|
| 10 |
+
|
| 11 |
+
# MIDIT — Checkpoints
|
| 12 |
+
|
| 13 |
+
Pretrained weights for **MIDIT**, a BERT-style masked language model for
|
| 14 |
+
symbolic music infilling (REMI+ tokenization, ~42.5M parameters).
|
| 15 |
+
|
| 16 |
+
Code, usage and full results: https://github.com/SimoneZanetti00/MIDIT
|
| 17 |
+
|
| 18 |
+
| File | Phase | Description |
|
| 19 |
+
|---|---|---|
|
| 20 |
+
| `midit-phase1-epoch14.ckpt` | Phase 1 (MLM pre-training) | base / zero-shot model, val_loss 2.7254 |
|
| 21 |
+
| `midit-infilling-epoch05.ckpt` | Phase 1b (BarFillMasking fine-tuning) | fine-tuned infilling model |
|
| 22 |
+
|
| 23 |
+
Weights contain the Lightning `state_dict` + `hyper_parameters` only
|
| 24 |
+
(optimizer state removed). Load with the model class from the repository.
|
| 25 |
+
|
| 26 |
+
Trained on the Lakh MIDI Dataset; released for research purposes.
|