---
license: mit
library_name: pytorch
tags:
- pytorch
- torch-geometric
- graph-neural-network
- reinforcement-learning
- transportation
- transit
- trndp
datasets:
- matrix-multiply/bloomington-tndp
arxiv: 2605.28730
---

# AlphaTransit Checkpoints

PyTorch checkpoints for **AlphaTransit: Learning to Design City-scale Transit Routes**.

## Files

| File | Setting | Bloomington service rate |
| --- | --- | ---: |
| `checkpoints/alphatransit_alpha0.3_best.pth` | `alpha=0.3`, `n_iter=500` | 54.64% |
| `checkpoints/alphatransit_alpha1.0_best.pth` | `alpha=1.0`, `n_iter=500` | 82.08% |

`model_kwargs.json` and `checkpoint_manifest.json` provide checkpoint metadata.

## Download

```python
from huggingface_hub import hf_hub_download

path = hf_hub_download(
    repo_id="matrix-multiply/alphatransit-checkpoints",
    filename="checkpoints/alphatransit_alpha0.3_best.pth",
)
```

## Links

- Paper: https://huggingface.co/papers/2605.28730
- Code: https://github.com/poudel-bibek/AlphaTransit
- Dataset: https://huggingface.co/datasets/matrix-multiply/bloomington-tndp
- Research preview: https://alphatransit.app/
- WandB report: https://api.wandb.ai/links/bibek-poudel/wa4zd1il