FM / transformers

/usr/local/lib/python3.6/site-packages/transformers/

/usr/local/lib/python3.6/site-packages/transformers

Name	Size	Mode	Actions
benchmark/	-	0755	rm
commands/	-	0755	rm
data/	-	0755	rm
models/	-	0755	rm
onnx/	-	0755	rm
pipelines/	-	0755	rm
sagemaker/	-	0755	rm
utils/	-	0755	rm
__pycache__/	-	0755	rm
activations.py	6612	0644	edit dl rm
activations_tf.py	4268	0644	edit dl rm
configuration_utils.py	46765	0644	edit dl rm
convert_graph_to_onnx.py	19575	0644	edit dl rm
convert_pytorch_checkpoint_to_tf2.py	16640	0644	edit dl rm
convert_slow_tokenizer.py	37626	0644	edit dl rm
convert_slow_tokenizers_checkpoints_to_fast.py	4954	0644	edit dl rm
convert_tf_hub_seq_to_seq_bert_to_pytorch.py	2899	0644	edit dl rm
debug_utils.py	12887	0644	edit dl rm
deepspeed.py	18883	0644	edit dl rm
dependency_versions_check.py	1767	0644	edit dl rm
dependency_versions_table.py	2390	0644	edit dl rm
dynamic_module_utils.py	19306	0644	edit dl rm
feature_extraction_sequence_utils.py	18074	0644	edit dl rm
feature_extraction_utils.py	26768	0644	edit dl rm
file_utils.py	3930	0644	edit dl rm
generation_beam_constraints.py	19078	0644	edit dl rm
generation_beam_search.py	40145	0644	edit dl rm
generation_flax_logits_process.py	10879	0644	edit dl rm
generation_flax_utils.py	38708	0644	edit dl rm
generation_logits_process.py	30336	0644	edit dl rm
generation_stopping_criteria.py	5509	0644	edit dl rm
generation_tf_logits_process.py	17503	0644	edit dl rm
generation_tf_utils.py	130671	0644	edit dl rm
generation_utils.py	177118	0644	edit dl rm
hf_argparser.py	11901	0644	edit dl rm
image_utils.py	12860	0644	edit dl rm
integrations.py	40203	0644	edit dl rm
keras_callbacks.py	18902	0644	edit dl rm
modelcard.py	35278	0644	edit dl rm
modeling_flax_outputs.py	35514	0644	edit dl rm
modeling_flax_pytorch_utils.py	12296	0644	edit dl rm
modeling_flax_utils.py	37661	0644	edit dl rm
modeling_outputs.py	60327	0644	edit dl rm
modeling_tf_outputs.py	46107	0644	edit dl rm
modeling_tf_pytorch_utils.py	19916	0644	edit dl rm
modeling_tf_utils.py	100832	0644	edit dl rm
modeling_utils.py	137800	0644	edit dl rm
optimization.py	27756	0644	edit dl rm
optimization_tf.py	15722	0644	edit dl rm
processing_utils.py	10559	0644	edit dl rm
py.typed	1	0644	edit dl rm
pytorch_utils.py	1649	0644	edit dl rm
testing_utils.py	48292	0644	edit dl rm
tf_utils.py	1568	0644	edit dl rm
tokenization_utils.py	39867	0644	edit dl rm
tokenization_utils_base.py	170159	0644	edit dl rm
tokenization_utils_fast.py	32909	0644	edit dl rm
trainer.py	143103	0644	edit dl rm
trainer_callback.py	23199	0644	edit dl rm
trainer_pt_utils.py	44764	0644	edit dl rm
trainer_seq2seq.py	10396	0644	edit dl rm
trainer_tf.py	34564	0644	edit dl rm
trainer_utils.py	18509	0644	edit dl rm
training_args.py	68042	0644	edit dl rm
training_args_seq2seq.py	2829	0644	edit dl rm
training_args_tf.py	14267	0644	edit dl rm
__init__.py	171952	0644	edit dl rm

Edit: /usr/local/lib/python3.6/site-packages/transformers/hf_argparser.py (11901B)

# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# 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.

import dataclasses
import json
import sys
from argparse import ArgumentDefaultsHelpFormatter, ArgumentParser, ArgumentTypeError
from copy import copy
from enum import Enum
from inspect import isclass
from pathlib import Path
from typing import Any, Dict, Iterable, NewType, Optional, Tuple, Union, get_type_hints

DataClass = NewType("DataClass", Any)
DataClassType = NewType("DataClassType", Any)

# From https://stackoverflow.com/questions/15008758/parsing-boolean-values-with-argparse
def string_to_bool(v):
    if isinstance(v, bool):
        return v
    if v.lower() in ("yes", "true", "t", "y", "1"):
        return True
    elif v.lower() in ("no", "false", "f", "n", "0"):
        return False
    else:
        raise ArgumentTypeError(
            f"Truthy value expected: got {v} but expected one of yes/no, true/false, t/f, y/n, 1/0 (case insensitive)."
        )

class HfArgumentParser(ArgumentParser):
    """
    This subclass of `argparse.ArgumentParser` uses type hints on dataclasses to generate arguments.

The class is designed to play well with the native argparse. In particular, you can add more (non-dataclass backed)
    arguments to the parser after initialization and you'll get the output back after parsing as an additional
    namespace. Optional: To create sub argument groups use the `_argument_group_name` attribute in the dataclass.
    """

dataclass_types: Iterable[DataClassType]

def __init__(self, dataclass_types: Union[DataClassType, Iterable[DataClassType]], **kwargs):
        """
        Args:
            dataclass_types:
                Dataclass type, or list of dataclass types for which we will "fill" instances with the parsed args.
            kwargs:
                (Optional) Passed to `argparse.ArgumentParser()` in the regular way.
        """
        # To make the default appear when using --help
        if "formatter_class" not in kwargs:
            kwargs["formatter_class"] = ArgumentDefaultsHelpFormatter
        super().__init__(**kwargs)
        if dataclasses.is_dataclass(dataclass_types):
            dataclass_types = [dataclass_types]
        self.dataclass_types = list(dataclass_types)
        for dtype in self.dataclass_types:
            self._add_dataclass_arguments(dtype)

@staticmethod
    def _parse_dataclass_field(parser: ArgumentParser, field: dataclasses.Field):
        field_name = f"--{field.name}"
        kwargs = field.metadata.copy()
        # field.metadata is not used at all by Data Classes,
        # it is provided as a third-party extension mechanism.
        if isinstance(field.type, str):
            raise RuntimeError(
                "Unresolved type detected, which should have been done with the help of "
                "`typing.get_type_hints` method by default"
            )

origin_type = getattr(field.type, "__origin__", field.type)
        if origin_type is Union:
            if len(field.type.__args__) != 2 or type(None) not in field.type.__args__:
                raise ValueError("Only `Union[X, NoneType]` (i.e., `Optional[X]`) is allowed for `Union`")
            if bool not in field.type.__args__:
                # filter `NoneType` in Union (except for `Union[bool, NoneType]`)
                field.type = (
                    field.type.__args__[0] if isinstance(None, field.type.__args__[1]) else field.type.__args__[1]
                )
                origin_type = getattr(field.type, "__origin__", field.type)

# A variable to store kwargs for a boolean field, if needed
        # so that we can init a `no_*` complement argument (see below)
        bool_kwargs = {}
        if isinstance(field.type, type) and issubclass(field.type, Enum):
            kwargs["choices"] = [x.value for x in field.type]
            kwargs["type"] = type(kwargs["choices"][0])
            if field.default is not dataclasses.MISSING:
                kwargs["default"] = field.default
            else:
                kwargs["required"] = True
        elif field.type is bool or field.type is Optional[bool]:
            # Copy the currect kwargs to use to instantiate a `no_*` complement argument below.
            # We do not initialize it here because the `no_*` alternative must be instantiated after the real argument
            bool_kwargs = copy(kwargs)

# Hack because type=bool in argparse does not behave as we want.
            kwargs["type"] = string_to_bool
            if field.type is bool or (field.default is not None and field.default is not dataclasses.MISSING):
                # Default value is False if we have no default when of type bool.
                default = False if field.default is dataclasses.MISSING else field.default
                # This is the value that will get picked if we don't include --field_name in any way
                kwargs["default"] = default
                # This tells argparse we accept 0 or 1 value after --field_name
                kwargs["nargs"] = "?"
                # This is the value that will get picked if we do --field_name (without value)
                kwargs["const"] = True
        elif isclass(origin_type) and issubclass(origin_type, list):
            kwargs["type"] = field.type.__args__[0]
            kwargs["nargs"] = "+"
            if field.default_factory is not dataclasses.MISSING:
                kwargs["default"] = field.default_factory()
            elif field.default is dataclasses.MISSING:
                kwargs["required"] = True
        else:
            kwargs["type"] = field.type
            if field.default is not dataclasses.MISSING:
                kwargs["default"] = field.default
            elif field.default_factory is not dataclasses.MISSING:
                kwargs["default"] = field.default_factory()
            else:
                kwargs["required"] = True
        parser.add_argument(field_name, **kwargs)

# Add a complement `no_*` argument for a boolean field AFTER the initial field has already been added.
        # Order is important for arguments with the same destination!
        # We use a copy of earlier kwargs because the original kwargs have changed a lot before reaching down
        # here and we do not need those changes/additional keys.
        if field.default is True and (field.type is bool or field.type is Optional[bool]):
            bool_kwargs["default"] = False
            parser.add_argument(f"--no_{field.name}", action="store_false", dest=field.name, **bool_kwargs)

def _add_dataclass_arguments(self, dtype: DataClassType):
        if hasattr(dtype, "_argument_group_name"):
            parser = self.add_argument_group(dtype._argument_group_name)
        else:
            parser = self

try:
            type_hints: Dict[str, type] = get_type_hints(dtype)
        except NameError:
            raise RuntimeError(
                f"Type resolution failed for f{dtype}. Try declaring the class in global scope or "
                f"removing line of `from __future__ import annotations` which opts in Postponed "
                f"Evaluation of Annotations (PEP 563)"
            )

for field in dataclasses.fields(dtype):
            if not field.init:
                continue
            field.type = type_hints[field.name]
            self._parse_dataclass_field(parser, field)

def parse_args_into_dataclasses(
        self, args=None, return_remaining_strings=False, look_for_args_file=True, args_filename=None
    ) -> Tuple[DataClass, ...]:
        """
        Parse command-line args into instances of the specified dataclass types.

This relies on argparse's `ArgumentParser.parse_known_args`. See the doc at:
        docs.python.org/3.7/library/argparse.html#argparse.ArgumentParser.parse_args

Args:
            args:
                List of strings to parse. The default is taken from sys.argv. (same as argparse.ArgumentParser)
            return_remaining_strings:
                If true, also return a list of remaining argument strings.
            look_for_args_file:
                If true, will look for a ".args" file with the same base name as the entry point script for this
                process, and will append its potential content to the command line args.
            args_filename:
                If not None, will uses this file instead of the ".args" file specified in the previous argument.

Returns:
            Tuple consisting of:

- the dataclass instances in the same order as they were passed to the initializer.abspath
                - if applicable, an additional namespace for more (non-dataclass backed) arguments added to the parser
                  after initialization.
                - The potential list of remaining argument strings. (same as argparse.ArgumentParser.parse_known_args)
        """
        if args_filename or (look_for_args_file and len(sys.argv)):
            if args_filename:
                args_file = Path(args_filename)
            else:
                args_file = Path(sys.argv[0]).with_suffix(".args")

if args_file.exists():
                fargs = args_file.read_text().split()
                args = fargs + args if args is not None else fargs + sys.argv[1:]
                # in case of duplicate arguments the first one has precedence
                # so we append rather than prepend.
        namespace, remaining_args = self.parse_known_args(args=args)
        outputs = []
        for dtype in self.dataclass_types:
            keys = {f.name for f in dataclasses.fields(dtype) if f.init}
            inputs = {k: v for k, v in vars(namespace).items() if k in keys}
            for k in keys:
                delattr(namespace, k)
            obj = dtype(**inputs)
            outputs.append(obj)
        if len(namespace.__dict__) > 0:
            # additional namespace.
            outputs.append(namespace)
        if return_remaining_strings:
            return (*outputs, remaining_args)
        else:
            if remaining_args:
                raise ValueError(f"Some specified arguments are not used by the HfArgumentParser: {remaining_args}")

return (*outputs,)

def parse_json_file(self, json_file: str) -> Tuple[DataClass, ...]:
        """
        Alternative helper method that does not use `argparse` at all, instead loading a json file and populating the
        dataclass types.
        """
        data = json.loads(Path(json_file).read_text())
        outputs = []
        for dtype in self.dataclass_types:
            keys = {f.name for f in dataclasses.fields(dtype) if f.init}
            inputs = {k: v for k, v in data.items() if k in keys}
            obj = dtype(**inputs)
            outputs.append(obj)
        return (*outputs,)

def parse_dict(self, args: dict) -> Tuple[DataClass, ...]:
        """
        Alternative helper method that does not use `argparse` at all, instead uses a dict and populating the dataclass
        types.
        """
        outputs = []
        for dtype in self.dataclass_types:
            keys = {f.name for f in dataclasses.fields(dtype) if f.init}
            inputs = {k: v for k, v in args.items() if k in keys}
            obj = dtype(**inputs)
            outputs.append(obj)
        return (*outputs,)

cmd: