def load_tf_weights_in_realm()

in src/transformers/models/deprecated/realm/modeling_realm.py [0:0]

• 89 lines of code
• 35 McCabe index (conditional complexity)

def load_tf_weights_in_realm(model, config, tf_checkpoint_path):
    """Load tf checkpoints in a pytorch model."""
    try:
        import re

        import numpy as np
        import tensorflow as tf
    except ImportError:
        logger.error(
            "Loading a TensorFlow model in PyTorch, requires TensorFlow to be installed. Please see "
            "https://www.tensorflow.org/install/ for installation instructions."
        )
        raise
    tf_path = os.path.abspath(tf_checkpoint_path)
    logger.info(f"Converting TensorFlow checkpoint from {tf_path}")
    # Load weights from TF model
    init_vars = tf.train.list_variables(tf_path)
    names = []
    arrays = []

    for name, shape in init_vars:
        logger.info(f"Loading TF weight {name} with shape {shape}")
        array = tf.train.load_variable(tf_path, name)
        names.append(name)
        arrays.append(array)

    for name, array in zip(names, arrays):
        if isinstance(model, RealmReader) and "reader" not in name:
            logger.info(f"Skipping {name} as it is not {model.__class__.__name__}'s parameter")
            continue

        # For pretrained openqa reader
        if (name.startswith("bert") or name.startswith("cls")) and isinstance(model, RealmForOpenQA):
            name = name.replace("bert/", "reader/realm/")
            name = name.replace("cls/", "reader/cls/")

        # For pretrained encoder
        if (name.startswith("bert") or name.startswith("cls")) and isinstance(model, RealmKnowledgeAugEncoder):
            name = name.replace("bert/", "realm/")

        # For finetuned reader
        if name.startswith("reader"):
            reader_prefix = "" if isinstance(model, RealmReader) else "reader/"
            name = name.replace("reader/module/bert/", f"{reader_prefix}realm/")
            name = name.replace("reader/module/cls/", f"{reader_prefix}cls/")
            name = name.replace("reader/dense/", f"{reader_prefix}qa_outputs/dense_intermediate/")
            name = name.replace("reader/dense_1/", f"{reader_prefix}qa_outputs/dense_output/")
            name = name.replace("reader/layer_normalization", f"{reader_prefix}qa_outputs/layer_normalization")

        # For embedder and scorer
        if name.startswith("module/module/module/"):  # finetuned
            embedder_prefix = "" if isinstance(model, RealmEmbedder) else "embedder/"
            name = name.replace("module/module/module/module/bert/", f"{embedder_prefix}realm/")
            name = name.replace("module/module/module/LayerNorm/", f"{embedder_prefix}cls/LayerNorm/")
            name = name.replace("module/module/module/dense/", f"{embedder_prefix}cls/dense/")
            name = name.replace("module/module/module/module/cls/predictions/", f"{embedder_prefix}cls/predictions/")
            name = name.replace("module/module/module/bert/", f"{embedder_prefix}realm/")
            name = name.replace("module/module/module/cls/predictions/", f"{embedder_prefix}cls/predictions/")
        elif name.startswith("module/module/"):  # pretrained
            embedder_prefix = "" if isinstance(model, RealmEmbedder) else "embedder/"
            name = name.replace("module/module/LayerNorm/", f"{embedder_prefix}cls/LayerNorm/")
            name = name.replace("module/module/dense/", f"{embedder_prefix}cls/dense/")

        name = name.split("/")
        # adam_v and adam_m are variables used in AdamWeightDecayOptimizer to calculated m and v
        # which are not required for using pretrained model
        if any(
            n in ["adam_v", "adam_m", "AdamWeightDecayOptimizer", "AdamWeightDecayOptimizer_1", "global_step"]
            for n in name
        ):
            logger.info(f"Skipping {'/'.join(name)}")
            continue
        pointer = model
        for m_name in name:
            if re.fullmatch(r"[A-Za-z]+_\d+", m_name):
                scope_names = re.split(r"_(\d+)", m_name)
            else:
                scope_names = [m_name]
            if scope_names[0] == "kernel" or scope_names[0] == "gamma":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "output_bias" or scope_names[0] == "beta":
                pointer = getattr(pointer, "bias")
            else:
                try:
                    pointer = getattr(pointer, scope_names[0])
                except AttributeError:
                    logger.info(f"Skipping {'/'.join(name)}")
                    continue
            if len(scope_names) >= 2:
                num = int(scope_names[1])
                pointer = pointer[num]
        if m_name[-11:] == "_embeddings":
            pointer = getattr(pointer, "weight")
        elif m_name == "kernel":
            array = np.transpose(array)
        try:
            assert pointer.shape == array.shape, (
                f"Pointer shape {pointer.shape} and array shape {array.shape} mismatched"
            )
        except AssertionError as e:
            e.args += (pointer.shape, array.shape)
            raise
        logger.info(f"Initialize PyTorch weight {name}")
        pointer.data = torch.from_numpy(array)
    return model