Possible bug in model initialization (`num_decoder_layers`)

[Snail-TI]

Possible bug in model initialization (num_decoder_layers)

Hello ProteinMPNN team,

First of all, thank you for releasing such an excellent tool and dataset — it has been very helpful in my work.

While reading through the code, I noticed what might be a small bug in the model initialization.
In the training script, the ProteinMPNN class is instantiated as follows:

model = ProteinMPNN(
    node_features=args.hidden_dim,
    edge_features=args.hidden_dim,
    hidden_dim=args.hidden_dim,
    num_encoder_layers=args.num_encoder_layers,
    num_decoder_layers=args.num_encoder_layers,   # <--- here
    k_neighbors=args.num_neighbors,
    dropout=args.dropout,
    augment_eps=args.backbone_noise,
)

Here, num_decoder_layers is set to args.num_encoder_layers.

However, based on the argument parser (--num_decoder_layers) defined in the repo, it seems the intended behavior should be:

num_decoder_layers=args.num_decoder_layers

This way, users can independently control the number of encoder and decoder layers.

Thanks again for making ProteinMPNN available to the community!