LabelAttentionClassifier.forward() creates label_indices on CPU, causing device mismatch errors when running on GPU.

Changes

• Create label_indices with explicit device and dtype matching token_embeddings:

# Before
label_indices = torch.arange(self.num_classes).expand(B, -1)

# After  
label_indices = torch.arange(
    self.num_classes, dtype=torch.long, device=token_embeddings.device
).expand(B, -1)

This ensures the embedding lookup operates on tensors from the same device, preventing runtime errors in GPU environments.

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.