# coding=utf-8
# Copyright 2020, The RAG Authors 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.
"""Tokenization classes for RAG."""
import os
import warnings
from contextlib import contextmanager
from typing import List, Optional
from ...tokenization_utils_base import BatchEncoding
from ...utils import logging
from .configuration_rag import RagConfig
logger = logging.get_logger(__name__)
[docs]class RagTokenizer:
def __init__(self, question_encoder, generator):
self.question_encoder = question_encoder
self.generator = generator
self.current_tokenizer = self.question_encoder
def save_pretrained(self, save_directory):
if os.path.isfile(save_directory):
raise ValueError(f"Provided path ({save_directory}) should be a directory, not a file")
os.makedirs(save_directory, exist_ok=True)
question_encoder_path = os.path.join(save_directory, "question_encoder_tokenizer")
generator_path = os.path.join(save_directory, "generator_tokenizer")
self.question_encoder.save_pretrained(question_encoder_path)
self.generator.save_pretrained(generator_path)
@classmethod
def from_pretrained(cls, pretrained_model_name_or_path, **kwargs):
# dynamically import AutoTokenizer
from ..auto.tokenization_auto import AutoTokenizer
config = kwargs.pop("config", None)
if config is None:
config = RagConfig.from_pretrained(pretrained_model_name_or_path)
question_encoder = AutoTokenizer.from_pretrained(
pretrained_model_name_or_path, config=config.question_encoder, subfolder="question_encoder_tokenizer"
)
generator = AutoTokenizer.from_pretrained(
pretrained_model_name_or_path, config=config.generator, subfolder="generator_tokenizer"
)
return cls(question_encoder=question_encoder, generator=generator)
def __call__(self, *args, **kwargs):
return self.current_tokenizer(*args, **kwargs)
def batch_decode(self, *args, **kwargs):
return self.generator.batch_decode(*args, **kwargs)
def decode(self, *args, **kwargs):
return self.generator.decode(*args, **kwargs)
[docs] @contextmanager
def as_target_tokenizer(self):
"""
Temporarily sets the tokenizer for encoding the targets. Useful for tokenizer associated to
sequence-to-sequence models that need a slightly different processing for the labels.
"""
self.current_tokenizer = self.generator
yield
self.current_tokenizer = self.question_encoder
def prepare_seq2seq_batch(
self,
src_texts: List[str],
tgt_texts: Optional[List[str]] = None,
max_length: Optional[int] = None,
max_target_length: Optional[int] = None,
padding: str = "longest",
return_tensors: str = None,
truncation: bool = True,
**kwargs,
) -> BatchEncoding:
warnings.warn(
"`prepare_seq2seq_batch` is deprecated and will be removed in version 5 of 🤗 Transformers. Use the "
"regular `__call__` method to prepare your inputs and the tokenizer under the `with_target_tokenizer` "
"context manager to prepare your targets. See the documentation of your specific tokenizer for more "
"details",
FutureWarning,
)
if max_length is None:
max_length = self.current_tokenizer.model_max_length
model_inputs = self(
src_texts,
add_special_tokens=True,
return_tensors=return_tensors,
max_length=max_length,
padding=padding,
truncation=truncation,
**kwargs,
)
if tgt_texts is None:
return model_inputs
# Process tgt_texts
with self.as_target_tokenizer():
if max_target_length is None:
max_target_length = self.current_tokenizer.model_max_length
labels = self(
tgt_texts,
add_special_tokens=True,
return_tensors=return_tensors,
padding=padding,
max_length=max_target_length,
truncation=truncation,
**kwargs,
)
model_inputs["labels"] = labels["input_ids"]
return model_inputs