[ENH] Add NNLossWrapper to support standard PyTorch nn.Module losses in pytorch-forecasting

[vinitjain2005]

Problem

Currently, base_metrics.py provides TorchMetricWrapper to wrap
torchmetrics Lightning metrics. However, there is no equivalent for
standard PyTorch nn.Module losses such as:

• torch.nn.HuberLoss
• torch.nn.MSELoss
• torch.nn.SmoothL1Loss
• torch.nn.L1Loss

If a user tries to pass these directly to a pytorch-forecasting model, it fails because they:

1. Don't implement update() / compute() required by LightningMetric
2. Don't handle the (target, weight) tuple that pytorch-forecasting
   passes internally (see MultiHorizonMetric.update() in base_metrics.py)
3. Don't implement to_prediction() which models call at inference time
4. Can't be composed with MultiLoss or CompositeMetric

Proposed Solution

Add NNLossWrapper in pytorch_forecasting/metrics/base_metrics.py as a sibling to the existing
TorchMetricWrapper, following the same
pattern:

class NNLossWrapper(Metric):
    """
    Wrap a standard PyTorch nn.Module loss for use with pytorch-forecasting.

    Example
    -------
    >>> loss = NNLossWrapper(nn.HuberLoss(delta=1.5))
    >>> combined = NNLossWrapper(nn.HuberLoss()) + MAE()
    """

    def __init__(self, loss_fn: torch.nn.Module, **kwargs):
        if not isinstance(loss_fn, torch.nn.Module):
            raise TypeError(f"loss_fn must be an nn.Module, got {type(loss_fn)}")
        super().__init__(**kwargs)
        self.loss_fn = loss_fn

    def update(self, y_pred: torch.Tensor, y_actual) -> None:
        if isinstance(y_actual, (list, tuple)) and not isinstance(
            y_actual, rnn.PackedSequence
        ):
            target, _ = y_actual
        else:
            target = y_actual
        y_pred = self.to_prediction(y_pred)
        self._loss = self.loss_fn(y_pred, target)

    def compute(self) -> torch.Tensor:
        return self._loss

    def to_prediction(self, y_pred: torch.Tensor) -> torch.Tensor:
        if y_pred.ndim == 3:
            y_pred = y_pred[..., 0]
        return y_pred

    def __repr__(self):
        return f"NNLossWrapper({repr(self.loss_fn)})"

Files to Change

• pytorch_forecasting/metrics/base_metrics.py — add NNLossWrapper
• pytorch_forecasting/metrics/__init__.py — export NNLossWrapper

Related

• Roadmap 2026 Roadmap 2026 #1993 — explicitly lists "Add adapters for nn losses"
• Existing TorchMetricWrapper in base_metrics.py — the parallel
  class this follows as a pattern

Github: @vinitjain2005