modeling_albert.py

# coding=utf-8
# Copyright 2018 Google AI, Google Brain and the HuggingFace Inc. team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""PyTorch ALBERT model. """

import math
import os
import warnings
from dataclasses import dataclass
from typing import Optional, Tuple

import torch
import torch.nn as nn
from torch.nn import CrossEntropyLoss, MSELoss

from .activations import ACT2FN
from .configuration_albert import AlbertConfig
from .file_utils import (
    ModelOutput,
    add_code_sample_docstrings,
    add_start_docstrings,
    add_start_docstrings_to_model_forward,
    replace_return_docstrings,
)
from .modeling_outputs import (
    BaseModelOutput,
    BaseModelOutputWithPooling,
    MaskedLMOutput,
    MultipleChoiceModelOutput,
    QuestionAnsweringModelOutput,
    SequenceClassifierOutput,
    TokenClassifierOutput,
)
from .modeling_utils import (
    PreTrainedModel,
    apply_chunking_to_forward,
    find_pruneable_heads_and_indices,
    prune_linear_layer,
)
from .utils import logging


logger = logging.get_logger(__name__)

_CONFIG_FOR_DOC = "AlbertConfig"
_TOKENIZER_FOR_DOC = "AlbertTokenizer"


ALBERT_PRETRAINED_MODEL_ARCHIVE_LIST = [
    "albert-base-v1",
    "albert-large-v1",
    "albert-xlarge-v1",
    "albert-xxlarge-v1",
    "albert-base-v2",
    "albert-large-v2",
    "albert-xlarge-v2",
    "albert-xxlarge-v2",
    # See all ALBERT models at https://huggingface.co/models?filter=albert
]


def load_tf_weights_in_albert(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("Converting TensorFlow checkpoint from {}".format(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("Loading TF weight {} with shape {}".format(name, shape))
        array = tf.train.load_variable(tf_path, name)
        names.append(name)
        arrays.append(array)

    for name, array in zip(names, arrays):
        print(name)

    for name, array in zip(names, arrays):
        original_name = name

        # If saved from the TF HUB module
        name = name.replace("module/", "")

        # Renaming and simplifying
        name = name.replace("ffn_1", "ffn")
        name = name.replace("bert/", "albert/")
        name = name.replace("attention_1", "attention")
        name = name.replace("transform/", "")
        name = name.replace("LayerNorm_1", "full_layer_layer_norm")
        name = name.replace("LayerNorm", "attention/LayerNorm")
        name = name.replace("transformer/", "")

        # The feed forward layer had an 'intermediate' step which has been abstracted away
        name = name.replace("intermediate/dense/", "")
        name = name.replace("ffn/intermediate/output/dense/", "ffn_output/")

        # ALBERT attention was split between self and output which have been abstracted away
        name = name.replace("/output/", "/")
        name = name.replace("/self/", "/")

        # The pooler is a linear layer
        name = name.replace("pooler/dense", "pooler")

        # The classifier was simplified to predictions from cls/predictions
        name = name.replace("cls/predictions", "predictions")
        name = name.replace("predictions/attention", "predictions")

        # Naming was changed to be more explicit
        name = name.replace("embeddings/attention", "embeddings")
        name = name.replace("inner_group_", "albert_layers/")
        name = name.replace("group_", "albert_layer_groups/")

        # Classifier
        if len(name.split("/")) == 1 and ("output_bias" in name or "output_weights" in name):
            name = "classifier/" + name

        # No ALBERT model currently handles the next sentence prediction task
        if "seq_relationship" in name:
            name = name.replace("seq_relationship/output_", "sop_classifier/classifier/")
            name = name.replace("weights", "weight")

        name = name.split("/")

        # Ignore the gradients applied by the LAMB/ADAM optimizers.
        if (
            "adam_m" in name
            or "adam_v" in name
            or "AdamWeightDecayOptimizer" in name
            or "AdamWeightDecayOptimizer_1" in name
            or "global_step" in name
        ):
            logger.info("Skipping {}".format("/".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")
            elif scope_names[0] == "output_weights":
                pointer = getattr(pointer, "weight")
            elif scope_names[0] == "squad":
                pointer = getattr(pointer, "classifier")
            else:
                try:
                    pointer = getattr(pointer, scope_names[0])
                except AttributeError:
                    logger.info("Skipping {}".format("/".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
        print("Initialize PyTorch weight {} from {}".format(name, original_name))
        pointer.data = torch.from_numpy(array)

    return model