File size: 1,734 Bytes
7ac8259 a3d83c2 7ac8259 a3d83c2 7ac8259 a3d83c2 7ac8259 a3d83c2 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 | ---
library_name: pytorch
license: other
license_name: exaone
license_link: https://huggingface.co/LG-AI-Research/CADET/blob/main/LICENSE
pipeline_tag: other
tags:
- reinforcement-learning
- imitation-learning
- pcb-routing
- kicad
- pcbworld
---
# CADET checkpoints
Paper: [CADET: Construction of PCB Routing Instances with Executable Demonstrations](https://huggingface.co/papers/2611.16913)
Code: [https://github.com/LGAI-Research/CADET](https://github.com/LGAI-Research/CADET)
Trained routing policies from the CADET paper. Weights only. All are decoder-only transformer policies
(6 layers, d_model 256, 8 heads, FFN 1024; 6.8M parameters), seed 42, run in
[PCBWorld](https://github.com/LGAI-Research/PCBWorld).
| File | Paper row | Training |
|---|---|---|
| `cadet.pt` | CADET | BC on RouteDemos, then PPO fine-tuning |
| `cadet_bc.pt` | CADET-BC | BC on RouteDemos |
| `ppo_routedemos.pt` | PPO (RouteDemos) | PPO from scratch on RouteDemos |
| `ppo_routedemos_no_tryroute.pt` | PPO (w/o TryRoute) | PPO from scratch on RouteDemos without TryRoute |
| `ppo_d2.pt` | PPO (D2) | PPO from scratch on PCBWorld D2 |
| `ppo_real.pt` | PPO (real) | PPO from scratch on D3-B ref (real boards) |
Each file holds the weights (`policy_state_dict`), the training arguments (`args`) and an observation probe
(`obs_probe`).
```bash
hf download LG-AI-Research/CADET --local-dir checkpoints
```
## License
EXAONE AI Model License Agreement 1.2 - NC ([LICENSE](https://huggingface.co/LG-AI-Research/CADET/blob/main/LICENSE)): research and educational use only.
Commercial use of the model and of its output, including boards routed with it, is not permitted.
The CADET and PCBWorld code is licensed separately; see their repositories. |