model_template.py

import torch
from torch import nn
from typing import Any, Iterable, List


class Model(nn.Module):
    """
    Template model for the leaderboard.

    Requirements:
    - Must be instantiable with no arguments (called by the evaluator).
    - Must implement `predict(batch)` which receives an iterable of inputs and
      returns a list of predictions (labels).
    - Must implement `eval()` to place the model in evaluation mode.
    - If you use PyTorch, submit a state_dict to be loaded via `load_state_dict`
    """

    def __init__(self, *args, **kwargs) -> None:
        super().__init__(*args, **kwargs)
        # Initialize your model here

    def eval(self) -> nn.Module:
        # Optional: set your model to evaluation mode
        return self

    def predict(self, batch: Iterable[Any]) -> List[Any]:
        """
        Implement your inference here.
        Inputs:
            batch: Iterable of preprocessed inputs (as produced by your preprocess.py)
        Returns:
            A list of predictions with the same length as `batch`.
        """
        raise NotImplementedError("Implement predict(...) to return a list of labels.")


def get_model() -> Model:
    """
    Factory function required by the evaluator.
    Returns an uninitialized model instance. The evaluator may optionally load
    weights (if provided) before calling predict(...).
    """
    return Model()