experiment: tfno-h1 device: cuda:1 seed: 42 model: CarstenNO model_config: n_modes: - 24 - 24 in_channels: 1 out_channels: 2 hidden_channels: 32 lifting_channels: 32 projection_channels: 32 rank: 0.5 norm: batch_norm n_layers: 10 domain_padding: 0.015625 domain_padding_mode: symmetric factorization: tucker implementation: factorized n_epochs: 101 loss: h1 loss_config: reductions: mean eval_losses: - l2 - h1 - grating_l2 - substrate_l2 - superstrate_l2 optim: AdamW optim_config: lr: 0.001 weight_decay: 1.0e-06 save_interval: 50 log_test_interval: 5 scheduler: OneCycleLR scheduler_config: max_lr: 0.01 total_steps: ${n_epochs} base_momentum: 0.85 max_momentum: 0.95 data: 1000-50-4