---
language:
- en
license: mit
library_name: pytorch
tags:
- music
- audio
- midi
- error-detection
- laddersym
---

# LadderSym Checkpoints (Official)

Official LadderSym checkpoints for reproducing the ICLR 2026 LadderSym results.

## Checkpoints

- `checkpoints/cocochorales/prompted/model.ckpt`
- `checkpoints/cocochorales/unprompted/model.ckpt`
- `checkpoints/maestro/prompted/model.ckpt`
- `checkpoints/maestro/unprompted/model.ckpt`



## Re-evaluated Checkpoint Results (March 2026)

The metrics below are from the latest re-evaluation runs for each official checkpoint.

Track mapping: `Track 0 = Extra`, `Track 1 = Missed`, `Track 2 = Correct`.

| Dataset | Variant | Onset F1 | Extra F1 (Track 0) | Missed F1 (Track 1) | Correct F1 (Track 2) |
|---|---|---:|---:|---:|---:|
| CocoChorales-E | Prompted | 0.9525 | 0.6007 | 0.5977 | 0.9712 |
| CocoChorales-E | Unprompted | 0.9488 | 0.5871 | 0.5923 | 0.9685 |
| MAESTRO-E | Prompted | 0.9389 | 0.8561 | 0.5349 | 0.9441 |
| MAESTRO-E | Unprompted | 0.9301 | 0.8002 | 0.4483 | 0.9349 |


## Download Example

```python
from huggingface_hub import hf_hub_download

ckpt = hf_hub_download(
    repo_id="ben2002chou/laddersym-checkpoints",
    filename="checkpoints/maestro/prompted/model.ckpt",
)
print(ckpt)
```

## Related Links

- Code: https://github.com/ben2002chou/LadderSYM
- Dataset (CocoChorales-E): https://huggingface.co/datasets/ben2002chou/CocoChorales-E
- Dataset (MAESTRO-E): https://huggingface.co/datasets/ben2002chou/MAESTRO-E
- Paper: https://openreview.net/forum?id=cizuvfyQXs