CADET checkpoints
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.
| 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).
hf download LG-AI-Research/CADET --local-dir checkpoints
License
EXAONE AI Model License Agreement 1.2 - NC (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.