Allow multiple pattern matches in chain with PyTorchFileRecorder

[laggui]

Pull Request Template

Checklist

• Confirmed that run-checks all script has been executed.

Changes

Changed the record remap function to allow multiple pattern matches in chain when using the PyTorchFileRecorder

For example, I initially though that providing multiple patterns like this would remap the keys on each match:

let load_args = LoadArgs::new("resnet18-f37072fd.pth".into())
  // Map *.downsample.0.* -> *.downsample.conv.*
  .with_key_remap("(.+)\\.downsample\\.0\\.(.+)", "$1.downsample.conv.$2")
  // Map *.downsample.1.* -> *.downsample.bn.*
  .with_key_remap("(.+)\\.downsample\\.1\\.(.+)", "$1.downsample.bn.$2")
  // Map layer[i].[j].* -> layer[i].blocks.[j].*
  .with_key_remap("layer[1-4]\\.([0-9])\\.(.+)", "layer$1.blocks.$2.$3");

let record = PyTorchFileRecorder::<FullPrecisionSettings>::new()
    .load(load_args, &device)
    .map_err(|err| format!("Failed to load weights.\nError: {err}"))
    .unwrap();

but it does not. Instead it breaks at the first match.

Stumbled upon this when trying to import the pre-trained ResNet-18 weights from torchvision for my implementation here.

In the residual blocks, we sometimes have a downsample layer that requires more than one portion of the key to be remapped (hence the patterns in chain). For example, layer1.0.downsample.0.* should be remapped to layer1.blocks.0.downsample.conv.*, which can be decomposed with the patterns presented in the example above.

[antimora]

LGTM