ptablasdepaula/complex-valued-modal-plate-transformer

DAFx Task B — Complex-Valued Transformer (leafy-wildflower-272, epoch 6)

Checkpoint for Task B (modal parameter estimation: frequency, decay, gain) of the 1st DAFx Parameter Estimation Challenge.

• Model: musicnet_complex_transformer (Eilers-style complex-valued Transformer encoder, warmup-cosine + grad-clip AdamW)
• wandb run: leafy-wildflower-272 (il3k0ipn)
• Epoch: 6 (resume continuation)
• Selection metric: `val/taskB_calibrated_RE 0.3328
• Checkpoint md5: e8b1d1967298875f6c473627a37bba49

Files

• leafy-wildflower-272_epoch006_RE0.3328.ckpt — PyTorch-Lightning checkpoint (model + optimizer state)
• config.yaml — Hydra config it was trained with (architecture + data + criterion)

Load

import torch
ckpt = torch.load("leafy-wildflower-272_epoch006_RE0.3328.ckpt", map_location="cpu", weights_only=False)

Rebuild the model from config.yaml (src.models.components.musicnet_complex_transformer.MusicNetComplexTransformer).