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.