s444391/UG-final
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

final ver

Browse Source
This commit is contained in:
Aleksandra Jonas 2023-02-14 23:44:20 +01:00
commit 69663af74a
13 changed files with 48517 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored Normal file
View File

@ -0,0 +1 @@
*/.DS_Store

6530
all_models.ipynb Normal file
View File

File diff suppressed because one or more lines are too long

2000
data/s2s-test.json Normal file
View File

File diff suppressed because it is too large Load Diff

16000
data/s2s-train.json Normal file
View File

File diff suppressed because it is too large Load Diff

2000
data/s2s-valid.json Normal file
View File

File diff suppressed because it is too large Load Diff

2000
data/test.json Normal file
View File

File diff suppressed because it is too large Load Diff

16000
data/train.json Normal file
View File

File diff suppressed because it is too large Load Diff

2000
data/valid.json Normal file
View File

File diff suppressed because it is too large Load Diff

79
data_prep.py Normal file
View File

@ -0,0 +1,79 @@
#!/usr/bin/env python3
import json
import logging
from pathlib import Path
from typing import List, Dict
from datasets import load_dataset
logger = logging.getLogger(__name__)
MAP_LABEL_TRANSLATION = {
0: 'sadness',
1: 'joy',
2: 'love',
3: 'anger',
4: 'fear',
5: 'surprise',
}
def save_as_translations(original_save_path: Path, data_to_save: List[Dict]) -> None:
file_name = 's2s-' + original_save_path.name
file_path = original_save_path.parent / file_name
print(f'Saving into: {file_path}')
with open(file_path, 'wt') as f_write:
for data_line in data_to_save:
label = data_line['label']
new_label = MAP_LABEL_TRANSLATION[label]
data_line['label'] = new_label
data_line_str = json.dumps(data_line)
f_write.write(f'{data_line_str}\n')
def main() -> None:
loaded_data = load_dataset('emotion')
logger.info(f'Loaded dataset emotion: {loaded_data}')
save_path = Path('data')
save_train_path = save_path / 'train.json'
save_valid_path = save_path / 'valid.json'
save_test_path = save_path / 'test.json'
if not save_path.exists():
save_path.mkdir()
data_train, data_valid, data_test = [], [], []
for source_data, dataset, max_size in [
(loaded_data['train'], data_train, None),
(loaded_data['test'], data_test, None),
(loaded_data['validation'], data_valid, None)
]:
for i, data in enumerate(source_data):
if max_size is not None and i >= max_size:
break
data_line = {
'label': int(data['label']),
'text': data['text'],
}
dataset.append(data_line)
logger.info(f'Train: {len(data_train):6d}')
logger.info(f'Test: {len(data_test):6d}')
logger.info(f'Validation: {len(data_valid):6d}')
for file_path, data_to_save in [
(save_train_path, data_train),
(save_valid_path, data_valid),
(save_test_path, data_test)
]:
print(f'Saving into: {file_path}')
with open(file_path, 'wt') as f_write:
for data_line in data_to_save:
data_line_str = json.dumps(data_line)
f_write.write(f'{data_line_str}\n')
save_as_translations(file_path, data_to_save)
if __name__ == '__main__':
main()

205
scripts/gpt2.py Normal file
View File

@ -0,0 +1,205 @@
import logging
from typing import Optional, Tuple, Union
import torch
from torch import nn
from torch.nn import MSELoss, CrossEntropyLoss, BCEWithLogitsLoss
from transformers import GPT2Model, GPT2ForSequenceClassification
from transformers.modeling_outputs import SequenceClassifierOutputWithPast
logger = logging.getLogger(__name__)
# Simple version #
class GPT2ClassificationHeadCustomSimple(nn.Module):
def __init__(self, config):
super().__init__()
hidden_size = config.n_embd
self.dense_1 = nn.Linear(hidden_size, 4 * hidden_size)
self.dense_2 = nn.Linear(4 * hidden_size, 2* hidden_size)
self.dense_3 = nn.Linear(2 * hidden_size, hidden_size)
self.dropout = nn.Dropout(config.resid_pdrop)
self.out_proj = nn.Linear(hidden_size, config.num_labels, bias=False)
def forward(self, x):
x = self.dense_1(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.dense_2(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.dense_3(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class GPT2ForSequenceClassificationCustomSimple(GPT2ForSequenceClassification):
_keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.transformer = GPT2Model(config)
self.score = GPT2ClassificationHeadCustomSimple(config)
# Model parallel
self.model_parallel = False
self.device_map = None
# Initialize weights and apply final processing
self.post_init()
# Version with custom forward 1 #
class GPT2ClassificationHeadCustom(nn.Module):
def __init__(self, config):
super().__init__()
hidden_size = config.n_embd
self.dense_1_input = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_1_hidden = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_2 = nn.Linear(4 * hidden_size, hidden_size)
self.dropout = nn.Dropout(config.resid_pdrop)
self.out_proj = nn.Linear(hidden_size, config.num_labels, bias=False)
def forward(self, x, **kwargs):
if 'hidden_states' in kwargs and kwargs['hidden_states'] is not None:
# Get hidden states from second from the end
hidden = kwargs['hidden_states'][-2]
else:
hidden = torch.zeros(x.size(), dtype=x.dtype, device=x.device)
x = self.dense_1_input(x)
x = torch.relu(x)
x = self.dropout(x)
hidden = self.dense_1_hidden(hidden)
hidden = torch.relu(hidden)
hidden = self.dropout(hidden)
x = torch.cat((x, hidden), dim=2)
x = self.dense_2(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class GPT2ForSequenceClassificationCustom(GPT2ForSequenceClassification):
_keys_to_ignore_on_load_missing = [r"h\.\d+\.attn\.masked_bias", r"lm_head.weight"]
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.transformer = GPT2Model(config)
self.score = GPT2ClassificationHeadCustom(config)
# Model parallel
self.model_parallel = False
self.device_map = None
# Initialize weights and apply final processing
self.post_init()
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
past_key_values: Optional[Tuple[Tuple[torch.Tensor]]] = None,
attention_mask: Optional[torch.FloatTensor] = None,
token_type_ids: Optional[torch.LongTensor] = None,
position_ids: Optional[torch.LongTensor] = None,
head_mask: Optional[torch.FloatTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
labels: Optional[torch.LongTensor] = None,
use_cache: Optional[bool] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple, SequenceClassifierOutputWithPast]:
r"""
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
"""
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
transformer_outputs = self.transformer(
input_ids,
past_key_values=past_key_values,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
use_cache=use_cache,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states or self.config.use_hidden_states,
return_dict=return_dict,
)
hidden_states = transformer_outputs[0]
logits = self.score(hidden_states, hidden_states=transformer_outputs.hidden_states)
if input_ids is not None:
batch_size, sequence_length = input_ids.shape[:2]
else:
batch_size, sequence_length = inputs_embeds.shape[:2]
assert (
self.config.pad_token_id is not None or batch_size == 1
), "Cannot handle batch sizes > 1 if no padding token is defined."
if self.config.pad_token_id is None:
sequence_lengths = -1
else:
if input_ids is not None:
sequence_lengths = torch.ne(input_ids, self.config.pad_token_id).sum(-1) - 1
else:
sequence_lengths = -1
logger.warning(
f"{self.__class__.__name__} will not detect padding tokens in `inputs_embeds`. Results may be "
"unexpected if using padding tokens in conjunction with `inputs_embeds.`"
)
pooled_logits = logits[torch.arange(batch_size, device=logits.device), sequence_lengths]
loss = None
if labels is not None:
if self.config.problem_type is None:
if self.num_labels == 1:
self.config.problem_type = "regression"
elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
self.config.problem_type = "single_label_classification"
else:
self.config.problem_type = "multi_label_classification"
if self.config.problem_type == "regression":
loss_fct = MSELoss()
if self.num_labels == 1:
loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
else:
loss = loss_fct(pooled_logits, labels)
elif self.config.problem_type == "single_label_classification":
loss_fct = CrossEntropyLoss()
loss = loss_fct(pooled_logits.view(-1, self.num_labels), labels.view(-1))
elif self.config.problem_type == "multi_label_classification":
loss_fct = BCEWithLogitsLoss()
loss = loss_fct(pooled_logits, labels)
if not return_dict:
output = (pooled_logits,) + transformer_outputs[1:]
return ((loss,) + output) if loss is not None else output
return SequenceClassifierOutputWithPast(
loss=loss,
logits=pooled_logits,
past_key_values=transformer_outputs.past_key_values,
hidden_states=transformer_outputs.hidden_states,
attentions=transformer_outputs.attentions,
)

334
scripts/roberta.py Normal file
View File

@ -0,0 +1,334 @@
from typing import Optional, Union, Tuple
import torch
from torch import nn
from torch.nn import MSELoss, CrossEntropyLoss, BCEWithLogitsLoss
from transformers import RobertaForSequenceClassification, RobertaModel
from transformers.modeling_outputs import SequenceClassifierOutput
# Simple version #
class RobertaClassificationHeadCustomSimple(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, config):
super().__init__()
hidden_size = config.hidden_size
self.dense_1 = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_1_hidden = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_2 = nn.Linear(2 * hidden_size, hidden_size)
classifier_dropout = (
config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
)
self.dropout = nn.Dropout(classifier_dropout)
self.relu = nn.LeakyReLU()
self.out_proj = nn.Linear(hidden_size, config.num_labels)
def forward(self, features, **kwargs):
x = features[:, 0, :] # take <s> token (equiv. to [CLS])
x = self.dense_1(x)
x = self.relu(x)
x = self.dropout(x)
hidden = self.dense_1_hidden(hidden)
hidden = torch.relu(hidden)
hidden = self.dropout(hidden)
x = torch.cat((x, hidden), dim=2)
x = self.dense_2(x)
x = self.relu(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class RobertaForSequenceClassificationCustomSimple(RobertaForSequenceClassification):
_keys_to_ignore_on_load_missing = [r"position_ids"]
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.config = config
self.roberta = RobertaModel(config, add_pooling_layer=False)
self.classifier = RobertaClassificationHeadCustomSimple(config)
# Initialize weights and apply final processing
self.init_weights()
# Version with custom forward 1 #
class RobertaClassificationHeadCustom(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, config):
super().__init__()
self.use_hidden_states = config.use_hidden_states
hidden_size = config.hidden_size
if self.use_hidden_states:
hidden_size *= 2
self.dense_1 = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_2 = nn.Linear(2 * hidden_size, hidden_size)
classifier_dropout = (
config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
)
self.dropout = nn.Dropout(classifier_dropout)
self.out_proj = nn.Linear(hidden_size, config.num_labels)
def forward(self, features, **kwargs):
if 'hidden_states' in kwargs and kwargs['hidden_states'] is not None:
if self.use_hidden_states:
x = torch.cat(
(
features[:, 0, :],
# take <s> token (equiv. to [CLS]) from hidden states from second from the end
kwargs['hidden_states'][-2][:, 0, :]
),
dim=1
)
else:
x = features[:, 0, :] + kwargs['hidden_states'][-2][:, 0, :]
del kwargs['hidden_states']
else:
x = features[:, 0, :] # take <s> token (equiv. to [CLS])
if self.use_hidden_states:
x = torch.cat(
(
features[:, 0, :],
torch.zeros(x.size(), dtype=x.dtype, device=x.device)
),
dim=1
)
x = self.dense_1(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.dense_2(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class RobertaForSequenceClassificationCustom(RobertaForSequenceClassification):
_keys_to_ignore_on_load_missing = [r"position_ids"]
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.config = config
self.roberta = RobertaModel(config, add_pooling_layer=False)
self.classifier = RobertaClassificationHeadCustom(config)
# Initialize weights and apply final processing
self.post_init()
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
attention_mask: Optional[torch.FloatTensor] = None,
token_type_ids: Optional[torch.LongTensor] = None,
position_ids: Optional[torch.LongTensor] = None,
head_mask: Optional[torch.FloatTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
labels: Optional[torch.LongTensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]:
r"""
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
"""
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.roberta(
input_ids,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states or self.config.use_hidden_states,
return_dict=return_dict,
)
sequence_output = outputs[0]
if return_dict:
logits = self.classifier(sequence_output, hidden_states=outputs.hidden_states)
else:
raise NotImplemented('Not implemented for using non-dictionary object')
loss = None
if labels is not None:
if self.config.problem_type is None:
if self.num_labels == 1:
self.config.problem_type = "regression"
elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
self.config.problem_type = "single_label_classification"
else:
self.config.problem_type = "multi_label_classification"
if self.config.problem_type == "regression":
loss_fct = MSELoss()
if self.num_labels == 1:
loss = loss_fct(logits.squeeze(), labels.squeeze())
else:
loss = loss_fct(logits, labels)
elif self.config.problem_type == "single_label_classification":
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
elif self.config.problem_type == "multi_label_classification":
loss_fct = BCEWithLogitsLoss()
loss = loss_fct(logits, labels)
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return SequenceClassifierOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)
# Version with custom forward 2 #
class RobertaClassificationHeadCustomAlternative(nn.Module):
"""Head for sentence-level classification tasks."""
def __init__(self, config):
super().__init__()
hidden_size = config.hidden_size
self.dense_1_input = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_1_hidden = nn.Linear(hidden_size, 2 * hidden_size)
self.dense_2 = nn.Linear(4 * hidden_size, hidden_size)
classifier_dropout = (
config.classifier_dropout if config.classifier_dropout is not None else config.hidden_dropout_prob
)
self.dropout = nn.Dropout(classifier_dropout)
self.out_proj = nn.Linear(hidden_size, config.num_labels)
def forward(self, features, **kwargs):
x = features[:, 0, :] # take <s> token (equiv. to [CLS])
if 'hidden_states' in kwargs and kwargs['hidden_states'] is not None:
# take <s> token (equiv. to [CLS]) from hidden states from second from the end
hidden = kwargs['hidden_states'][-2][:, 0, :]
else:
hidden = torch.zeros(x.size(), dtype=x.dtype, device=x.device)
x = self.dense_1_input(x)
x = torch.relu(x)
x = self.dropout(x)
hidden = self.dense_1_hidden(hidden)
hidden = torch.relu(hidden)
hidden = self.dropout(hidden)
x = torch.cat((x, hidden), dim=1)
x = self.dense_2(x)
x = torch.relu(x)
x = self.dropout(x)
x = self.out_proj(x)
return x
class RobertaForSequenceClassificationCustomAlternative(RobertaForSequenceClassification):
_keys_to_ignore_on_load_missing = [r"position_ids"]
def __init__(self, config):
super().__init__(config)
self.num_labels = config.num_labels
self.config = config
self.roberta = RobertaModel(config, add_pooling_layer=False)
self.classifier = RobertaClassificationHeadCustomAlternative(config)
self.init_weights()
def forward(
self,
input_ids: Optional[torch.LongTensor] = None,
attention_mask: Optional[torch.FloatTensor] = None,
token_type_ids: Optional[torch.LongTensor] = None,
position_ids: Optional[torch.LongTensor] = None,
head_mask: Optional[torch.FloatTensor] = None,
inputs_embeds: Optional[torch.FloatTensor] = None,
labels: Optional[torch.LongTensor] = None,
output_attentions: Optional[bool] = None,
output_hidden_states: Optional[bool] = None,
return_dict: Optional[bool] = None,
) -> Union[Tuple[torch.Tensor], SequenceClassifierOutput]:
r"""
labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
`config.num_labels > 1` a classification loss is computed (Cross-Entropy).
"""
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
outputs = self.roberta(
input_ids,
attention_mask=attention_mask,
token_type_ids=token_type_ids,
position_ids=position_ids,
head_mask=head_mask,
inputs_embeds=inputs_embeds,
output_attentions=output_attentions,
output_hidden_states=output_hidden_states or self.config.use_hidden_states,
return_dict=return_dict,
)
sequence_output = outputs[0]
if return_dict:
logits = self.classifier(sequence_output, hidden_states=outputs.hidden_states)
else:
raise NotImplemented('Not implemented for using non-dictionary object')
loss = None
if labels is not None:
if self.config.problem_type is None:
if self.num_labels == 1:
self.config.problem_type = "regression"
elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
self.config.problem_type = "single_label_classification"
else:
self.config.problem_type = "multi_label_classification"
if self.config.problem_type == "regression":
loss_fct = MSELoss()
if self.num_labels == 1:
loss = loss_fct(logits.squeeze(), labels.squeeze())
else:
loss = loss_fct(logits, labels)
elif self.config.problem_type == "single_label_classification":
loss_fct = CrossEntropyLoss()
loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
elif self.config.problem_type == "multi_label_classification":
loss_fct = BCEWithLogitsLoss()
loss = loss_fct(logits, labels)
if not return_dict:
output = (logits,) + outputs[2:]
return ((loss,) + output) if loss is not None else output
return SequenceClassifierOutput(
loss=loss,
logits=logits,
hidden_states=outputs.hidden_states,
attentions=outputs.attentions,
)

661
scripts/run_glue.py Normal file
View File

@ -0,0 +1,661 @@
#!/usr/bin/env python
# coding=utf-8
# Copyright 2020 The HuggingFace Inc. 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.
""" Finetuning the library models for sequence classification on GLUE."""
# You can also adapt this script on your own text classification task. Pointers for this are left as comments.
import logging
import os
import random
import sys
from dataclasses import dataclass, field
from typing import Optional
import datasets
import numpy as np
from datasets import load_dataset
import evaluate
import transformers
from transformers import (
AutoConfig,
AutoModelForSequenceClassification,
AutoTokenizer,
DataCollatorWithPadding,
EvalPrediction,
HfArgumentParser,
PretrainedConfig,
Trainer,
TrainingArguments,
default_data_collator,
set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
from gpt2 import GPT2ForSequenceClassificationCustom
from roberta import RobertaForSequenceClassificationCustomAlternative
MODEL_NAME_TO_CLASS = {
'roberta_custom': RobertaForSequenceClassificationCustomAlternative,
'gpt2_custom': GPT2ForSequenceClassificationCustom
}
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.23.0")
require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/text-classification/requirements.txt")
task_to_keys = {
"cola": ("sentence", None),
"mnli": ("premise", "hypothesis"),
"mrpc": ("sentence1", "sentence2"),
"qnli": ("question", "sentence"),
"qqp": ("question1", "question2"),
"rte": ("sentence1", "sentence2"),
"sst2": ("sentence", None),
"stsb": ("sentence1", "sentence2"),
"wnli": ("sentence1", "sentence2"),
}
logger = logging.getLogger(__name__)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
Using `HfArgumentParser` we can turn this class
into argparse arguments to be able to specify them on
the command line.
"""
task_name: Optional[str] = field(
default=None,
metadata={"help": "The name of the task to train on: " + ", ".join(task_to_keys.keys())},
)
dataset_name: Optional[str] = field(
default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
)
dataset_config_name: Optional[str] = field(
default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
)
max_seq_length: int = field(
default=128,
metadata={
"help": (
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded."
)
},
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached preprocessed datasets or not."}
)
pad_to_max_length: bool = field(
default=True,
metadata={
"help": (
"Whether to pad all samples to `max_seq_length`. "
"If False, will pad the samples dynamically when batching to the maximum length in the batch."
)
},
)
max_train_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of training examples to this "
"value if set."
)
},
)
max_eval_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of evaluation examples to this "
"value if set."
)
},
)
max_predict_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of prediction examples to this "
"value if set."
)
},
)
train_file: Optional[str] = field(
default=None, metadata={"help": "A csv or a json file containing the training data."}
)
validation_file: Optional[str] = field(
default=None, metadata={"help": "A csv or a json file containing the validation data."}
)
test_file: Optional[str] = field(default=None, metadata={"help": "A csv or a json file containing the test data."})
def __post_init__(self):
if self.task_name is not None:
self.task_name = self.task_name.lower()
if self.task_name not in task_to_keys.keys():
raise ValueError("Unknown task, you should pick one in " + ",".join(task_to_keys.keys()))
elif self.dataset_name is not None:
pass
elif self.train_file is None or self.validation_file is None:
raise ValueError("Need either a GLUE task, a training/validation file or a dataset name.")
else:
train_extension = self.train_file.split(".")[-1]
assert train_extension in ["csv", "json"], "`train_file` should be a csv or a json file."
validation_extension = self.validation_file.split(".")[-1]
assert (
validation_extension == train_extension
), "`validation_file` should have the same extension (csv or json) as `train_file`."
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
"""
model_name_or_path: str = field(
metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where do you want to store the pretrained models downloaded from huggingface.co"},
)
use_fast_tokenizer: bool = field(
default=True,
metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
)
model_revision: str = field(
default="main",
metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
)
use_auth_token: bool = field(
default=False,
metadata={
"help": (
"Will use the token generated when running `huggingface-cli login` (necessary to use this script "
"with private models)."
)
},
)
ignore_mismatched_sizes: bool = field(
default=False,
metadata={"help": "Will enable to load a pretrained model whose head dimensions are different."},
)
custom_model: str = field(
default=None,
metadata={
"help": "Use custom implementation from available list",
"choices": list(MODEL_NAME_TO_CLASS.keys()),
},
)
def main():
# See all possible arguments in src/transformers/training_args.py
# or by passing the --help flag to this script.
# We now keep distinct sets of args, for a cleaner separation of concerns.
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, TrainingArguments))
if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
# If we pass only one argument to the script and it's the path to a json file,
# let's parse it to get our arguments.
model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
else:
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
# Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
# information sent is the one passed as arguments along with your Python/PyTorch versions.
send_example_telemetry("run_glue", model_args, data_args)
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
log_level = training_args.get_process_log_level()
logger.setLevel(log_level)
datasets.utils.logging.set_verbosity(log_level)
transformers.utils.logging.set_verbosity(log_level)
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+ f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
logger.info(f"Training/evaluation parameters {training_args}")
# Detecting last checkpoint.
last_checkpoint = None
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
last_checkpoint = get_last_checkpoint(training_args.output_dir)
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
raise ValueError(
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
"Use --overwrite_output_dir to overcome."
)
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
logger.info(
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
)
# Set seed before initializing model.
set_seed(training_args.seed)
# Get the datasets: you can either provide your own CSV/JSON training and evaluation files (see below)
# or specify a GLUE benchmark task (the dataset will be downloaded automatically from the datasets Hub).
#
# For CSV/JSON files, this script will use as labels the column called 'label' and as pair of sentences the
# sentences in columns called 'sentence1' and 'sentence2' if such column exists or the first two columns not named
# label if at least two columns are provided.
#
# If the CSVs/JSONs contain only one non-label column, the script does single sentence classification on this
# single column. You can easily tweak this behavior (see below)
#
# In distributed training, the load_dataset function guarantee that only one local process can concurrently
# download the dataset.
if data_args.task_name is not None:
# Downloading and loading a dataset from the hub.
raw_datasets = load_dataset(
"glue",
data_args.task_name,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
elif data_args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
raw_datasets = load_dataset(
data_args.dataset_name,
data_args.dataset_config_name,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
else:
# Loading a dataset from your local files.
# CSV/JSON training and evaluation files are needed.
data_files = {"train": data_args.train_file, "validation": data_args.validation_file}
# Get the test dataset: you can provide your own CSV/JSON test file (see below)
# when you use `do_predict` without specifying a GLUE benchmark task.
if training_args.do_predict:
if data_args.test_file is not None:
train_extension = data_args.train_file.split(".")[-1]
test_extension = data_args.test_file.split(".")[-1]
assert (
test_extension == train_extension
), "`test_file` should have the same extension (csv or json) as `train_file`."
data_files["test"] = data_args.test_file
else:
raise ValueError("Need either a GLUE task or a test file for `do_predict`.")
for key in data_files.keys():
logger.info(f"load a local file for {key}: {data_files[key]}")
if data_args.train_file.endswith(".csv"):
# Loading a dataset from local csv files
raw_datasets = load_dataset(
"csv",
data_files=data_files,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
else:
# Loading a dataset from local json files
raw_datasets = load_dataset(
"json",
data_files=data_files,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
# See more about loading any type of standard or custom dataset at
# https://huggingface.co/docs/datasets/loading_datasets.html.
# Labels
if data_args.task_name is not None:
is_regression = data_args.task_name == "stsb"
if not is_regression:
label_list = raw_datasets["train"].features["label"].names
num_labels = len(label_list)
else:
num_labels = 1
else:
# Trying to have good defaults here, don't hesitate to tweak to your needs.
is_regression = raw_datasets["train"].features["label"].dtype in ["float32", "float64"]
if is_regression:
num_labels = 1
else:
# A useful fast method:
# https://huggingface.co/docs/datasets/package_reference/main_classes.html#datasets.Dataset.unique
label_list = raw_datasets["train"].unique("label")
label_list.sort() # Let's sort it for determinism
num_labels = len(label_list)
# Load pretrained model and tokenizer
#
# In distributed training, the .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
config = AutoConfig.from_pretrained(
model_args.config_name if model_args.config_name else model_args.model_name_or_path,
num_labels=num_labels,
finetuning_task=data_args.task_name,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
tokenizer = AutoTokenizer.from_pretrained(
model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
cache_dir=model_args.cache_dir,
use_fast=model_args.use_fast_tokenizer,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
custom_model = model_args.custom_model
if custom_model is not None:
if 'roberta' in custom_model and 'roberta' not in model_args.model_name_or_path:
raise RuntimeError('Model and custom implementation should be the same type: RoBERTa')
elif 'gpt2' in custom_model and 'gpt2' not in model_args.model_name_or_path:
raise RuntimeError('Model and custom implementation should be the same type: GPT-2')
config.use_hidden_states = 'hidden' in custom_model
logger.info(f'Using hidden states in model: {config.use_hidden_states}')
model_cls = MODEL_NAME_TO_CLASS[custom_model]
else:
model_cls = AutoModelForSequenceClassification
logger.info(f'Using implementation from class: {model_cls.__name__}')
model = model_cls.from_pretrained(
model_args.model_name_or_path,
from_tf=bool(".ckpt" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
ignore_mismatched_sizes=model_args.ignore_mismatched_sizes,
)
if model_args.model_name_or_path.startswith('roberta'):
for name, param in model.named_parameters():
if name.startswith('roberta.encoder.layer'):
layer_number = name.split('.')[3]
if int(layer_number) % 2 == 0:
param.requires_grad = False
frozen_layers=[(name,param.requires_grad) for (name, param) in model.named_parameters() if not param.requires_grad]
print('\n\nFrozen layers:')
print(frozen_layers,'\n\n')
if 'gpt2' in tokenizer.name_or_path and tokenizer.pad_token is None:
logger.info(f'Set PAD token to EOS: {tokenizer.eos_token}')
tokenizer._pad_token = tokenizer.eos_token
model.config.pad_token_id = model.config.eos_token_id
# Preprocessing the raw_datasets
if data_args.task_name is not None:
sentence1_key, sentence2_key = task_to_keys[data_args.task_name]
else:
# Again, we try to have some nice defaults but don't hesitate to tweak to your use case.
non_label_column_names = [name for name in raw_datasets["train"].column_names if name != "label"]
if "sentence1" in non_label_column_names and "sentence2" in non_label_column_names:
sentence1_key, sentence2_key = "sentence1", "sentence2"
else:
if len(non_label_column_names) >= 2:
sentence1_key, sentence2_key = non_label_column_names[:2]
else:
sentence1_key, sentence2_key = non_label_column_names[0], None
# Padding strategy
if data_args.pad_to_max_length:
padding = "max_length"
else:
# We will pad later, dynamically at batch creation, to the max sequence length in each batch
padding = False
# Some models have set the order of the labels to use, so let's make sure we do use it.
label_to_id = None
if (
model.config.label2id != PretrainedConfig(num_labels=num_labels).label2id
and data_args.task_name is not None
and not is_regression
):
# Some have all caps in their config, some don't.
label_name_to_id = {k.lower(): v for k, v in model.config.label2id.items()}
if list(sorted(label_name_to_id.keys())) == list(sorted(label_list)):
label_to_id = {i: int(label_name_to_id[label_list[i]]) for i in range(num_labels)}
else:
logger.warning(
"Your model seems to have been trained with labels, but they don't match the dataset: ",
f"model labels: {list(sorted(label_name_to_id.keys()))}, dataset labels: {list(sorted(label_list))}."
"\nIgnoring the model labels as a result.",
)
elif data_args.task_name is None and not is_regression:
label_to_id = {v: i for i, v in enumerate(label_list)}
if label_to_id is not None:
model.config.label2id = label_to_id
model.config.id2label = {id: label for label, id in config.label2id.items()}
elif data_args.task_name is not None and not is_regression:
model.config.label2id = {l: i for i, l in enumerate(label_list)}
model.config.id2label = {id: label for label, id in config.label2id.items()}
if data_args.max_seq_length > tokenizer.model_max_length:
logger.warning(
f"The max_seq_length passed ({data_args.max_seq_length}) is larger than the maximum length for the"
f"model ({tokenizer.model_max_length}). Using max_seq_length={tokenizer.model_max_length}."
)
max_seq_length = min(data_args.max_seq_length, tokenizer.model_max_length)
def preprocess_function(examples):
# Tokenize the texts
args = (
(examples[sentence1_key],) if sentence2_key is None else (examples[sentence1_key], examples[sentence2_key])
)
result = tokenizer(*args, padding=padding, max_length=max_seq_length, truncation=True)
# Map labels to IDs (not necessary for GLUE tasks)
if label_to_id is not None and "label" in examples:
result["label"] = [(label_to_id[l] if l != -1 else -1) for l in examples["label"]]
return result
with training_args.main_process_first(desc="dataset map pre-processing"):
raw_datasets = raw_datasets.map(
preprocess_function,
batched=True,
load_from_cache_file=not data_args.overwrite_cache,
desc="Running tokenizer on dataset",
)
if training_args.do_train:
if "train" not in raw_datasets:
raise ValueError("--do_train requires a train dataset")
train_dataset = raw_datasets["train"]
if data_args.max_train_samples is not None:
max_train_samples = min(len(train_dataset), data_args.max_train_samples)
train_dataset = train_dataset.select(range(max_train_samples))
if training_args.do_eval:
if "validation" not in raw_datasets and "validation_matched" not in raw_datasets:
raise ValueError("--do_eval requires a validation dataset")
eval_dataset = raw_datasets["validation_matched" if data_args.task_name == "mnli" else "validation"]
if data_args.max_eval_samples is not None:
max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
eval_dataset = eval_dataset.select(range(max_eval_samples))
if training_args.do_predict or data_args.task_name is not None or data_args.test_file is not None:
if "test" not in raw_datasets and "test_matched" not in raw_datasets:
raise ValueError("--do_predict requires a test dataset")
predict_dataset = raw_datasets["test_matched" if data_args.task_name == "mnli" else "test"]
if data_args.max_predict_samples is not None:
max_predict_samples = min(len(predict_dataset), data_args.max_predict_samples)
predict_dataset = predict_dataset.select(range(max_predict_samples))
# Log a few random samples from the training set:
if training_args.do_train:
for index in random.sample(range(len(train_dataset)), 3):
logger.info(f"Sample {index} of the training set: {train_dataset[index]}.")
# Get the metric function
if data_args.task_name is not None:
metric = evaluate.load("glue", data_args.task_name)
else:
metric = evaluate.load("accuracy")
# You can define your custom compute_metrics function. It takes an `EvalPrediction` object (a namedtuple with a
# predictions and label_ids field) and has to return a dictionary string to float.
def compute_metrics(p: EvalPrediction):
preds = p.predictions[0] if isinstance(p.predictions, tuple) else p.predictions
preds = np.squeeze(preds) if is_regression else np.argmax(preds, axis=1)
if data_args.task_name is not None:
result = metric.compute(predictions=preds, references=p.label_ids)
if len(result) > 1:
result["combined_score"] = np.mean(list(result.values())).item()
return result
elif is_regression:
return {"mse": ((preds - p.label_ids) ** 2).mean().item()}
else:
return {"accuracy": (preds == p.label_ids).astype(np.float32).mean().item()}
# Data collator will default to DataCollatorWithPadding when the tokenizer is passed to Trainer, so we change it if
# we already did the padding.
if data_args.pad_to_max_length:
data_collator = default_data_collator
elif training_args.fp16:
data_collator = DataCollatorWithPadding(tokenizer, pad_to_multiple_of=8)
else:
data_collator = None
# Initialize our Trainer
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset if training_args.do_train else None,
eval_dataset=eval_dataset if training_args.do_eval else None,
compute_metrics=compute_metrics,
tokenizer=tokenizer,
data_collator=data_collator,
)
# Training
if training_args.do_train:
checkpoint = None
if training_args.resume_from_checkpoint is not None:
checkpoint = training_args.resume_from_checkpoint
elif last_checkpoint is not None:
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
metrics = train_result.metrics
max_train_samples = (
data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
)
metrics["train_samples"] = min(max_train_samples, len(train_dataset))
trainer.save_model() # Saves the tokenizer too for easy upload
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()
# Evaluation
if training_args.do_eval:
logger.info("*** Evaluate ***")
# Loop to handle MNLI double evaluation (matched, mis-matched)
tasks = [data_args.task_name]
eval_datasets = [eval_dataset]
if data_args.task_name == "mnli":
tasks.append("mnli-mm")
valid_mm_dataset = raw_datasets["validation_mismatched"]
if data_args.max_eval_samples is not None:
max_eval_samples = min(len(valid_mm_dataset), data_args.max_eval_samples)
valid_mm_dataset = valid_mm_dataset.select(range(max_eval_samples))
eval_datasets.append(valid_mm_dataset)
combined = {}
for eval_dataset, task in zip(eval_datasets, tasks):
metrics = trainer.evaluate(eval_dataset=eval_dataset)
max_eval_samples = (
data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
)
metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
if task == "mnli-mm":
metrics = {k + "_mm": v for k, v in metrics.items()}
if task is not None and "mnli" in task:
combined.update(metrics)
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", combined if task is not None and "mnli" in task else metrics)
if training_args.do_predict:
logger.info("*** Predict ***")
# Loop to handle MNLI double evaluation (matched, mis-matched)
tasks = [data_args.task_name]
predict_datasets = [predict_dataset]
if data_args.task_name == "mnli":
tasks.append("mnli-mm")
predict_datasets.append(raw_datasets["test_mismatched"])
for predict_dataset, task in zip(predict_datasets, tasks):
# Removing the `label` columns because it contains -1 and Trainer won't like that.
predict_dataset = predict_dataset.remove_columns("label")
predictions = trainer.predict(predict_dataset, metric_key_prefix="predict").predictions
predictions = np.squeeze(predictions) if is_regression else np.argmax(predictions, axis=1)
output_predict_file = os.path.join(training_args.output_dir, f"predict_results_{task}.txt")
if trainer.is_world_process_zero():
with open(output_predict_file, "w") as writer:
logger.info(f"***** Predict results {task} *****")
writer.write("index\tprediction\n")
for index, item in enumerate(predictions):
if is_regression:
writer.write(f"{index}\t{item:3.3f}\n")
else:
item = label_list[item]
writer.write(f"{index}\t{item}\n")
kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "text-classification"}
if data_args.task_name is not None:
kwargs["language"] = "en"
kwargs["dataset_tags"] = "glue"
kwargs["dataset_args"] = data_args.task_name
kwargs["dataset"] = f"GLUE {data_args.task_name.upper()}"
if training_args.push_to_hub:
trainer.push_to_hub(**kwargs)
else:
trainer.create_model_card(**kwargs)
def _mp_fn(index):
# For xla_spawn (TPUs)
main()
if __name__ == "__main__":
main()

707
scripts/run_translation.py Normal file
View File

@ -0,0 +1,707 @@
#!/usr/bin/env python
# coding=utf-8
# Copyright The HuggingFace Team and The HuggingFace Inc. 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.
"""
Fine-tuning the library models for sequence to sequence.
"""
# You can also adapt this script on your own sequence to sequence task. Pointers for this are left as comments.
from collections import defaultdict
import logging
import os
import sys
from dataclasses import dataclass, field
from typing import Optional
import torch
import datasets
import numpy as np
from datasets import load_dataset
import evaluate
import transformers
from transformers import (
AutoConfig,
AutoModelForSeq2SeqLM,
AutoTokenizer,
DataCollatorForSeq2Seq,
HfArgumentParser,
M2M100Tokenizer,
MBart50Tokenizer,
MBart50TokenizerFast,
MBartTokenizer,
MBartTokenizerFast,
Seq2SeqTrainer,
Seq2SeqTrainingArguments,
default_data_collator,
set_seed,
)
from transformers.trainer_utils import get_last_checkpoint
from transformers.utils import check_min_version, send_example_telemetry
from transformers.utils.versions import require_version
# Will error if the minimal version of Transformers is not installed. Remove at your own risks.
check_min_version("4.23.0")
require_version("datasets>=1.8.0", "To fix: pip install -r examples/pytorch/translation/requirements.txt")
logger = logging.getLogger(__name__)
# A list of all multilingual tokenizer which require src_lang and tgt_lang attributes.
MULTILINGUAL_TOKENIZERS = [MBartTokenizer, MBartTokenizerFast, MBart50Tokenizer, MBart50TokenizerFast, M2M100Tokenizer]
MAP_LABEL_TRANSLATION = {
0: 'sadness',
1: 'joy',
2: 'love',
3: 'anger',
4: 'fear',
5: 'surprise',
}
@dataclass
class ModelArguments:
"""
Arguments pertaining to which model/config/tokenizer we are going to fine-tune from.
"""
model_name_or_path: str = field(
metadata={"help": "Path to pretrained model or model identifier from huggingface.co/models"}
)
config_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained config name or path if not the same as model_name"}
)
tokenizer_name: Optional[str] = field(
default=None, metadata={"help": "Pretrained tokenizer name or path if not the same as model_name"}
)
cache_dir: Optional[str] = field(
default=None,
metadata={"help": "Where to store the pretrained models downloaded from huggingface.co"},
)
use_fast_tokenizer: bool = field(
default=True,
metadata={"help": "Whether to use one of the fast tokenizer (backed by the tokenizers library) or not."},
)
model_revision: str = field(
default="main",
metadata={"help": "The specific model version to use (can be a branch name, tag name or commit id)."},
)
use_auth_token: bool = field(
default=False,
metadata={
"help": (
"Will use the token generated when running `huggingface-cli login` (necessary to use this script "
"with private models)."
)
},
)
freeze_weights: bool = field(
default=False,
metadata={"help": "Freeze encoder weights"},
)
@dataclass
class DataTrainingArguments:
"""
Arguments pertaining to what data we are going to input our model for training and eval.
"""
source_lang: str = field(default=None, metadata={"help": "Source language id for translation."})
target_lang: str = field(default=None, metadata={"help": "Target language id for translation."})
dataset_name: Optional[str] = field(
default=None, metadata={"help": "The name of the dataset to use (via the datasets library)."}
)
dataset_config_name: Optional[str] = field(
default=None, metadata={"help": "The configuration name of the dataset to use (via the datasets library)."}
)
train_file: Optional[str] = field(default=None, metadata={"help": "The input training data file (a jsonlines)."})
validation_file: Optional[str] = field(
default=None,
metadata={
"help": "An optional input evaluation data file to evaluate the metrics (sacrebleu) on a jsonlines file."
},
)
test_file: Optional[str] = field(
default=None,
metadata={"help": "An optional input test data file to evaluate the metrics (sacrebleu) on a jsonlines file."},
)
overwrite_cache: bool = field(
default=False, metadata={"help": "Overwrite the cached training and evaluation sets"}
)
preprocessing_num_workers: Optional[int] = field(
default=None,
metadata={"help": "The number of processes to use for the preprocessing."},
)
max_source_length: Optional[int] = field(
default=1024,
metadata={
"help": (
"The maximum total input sequence length after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded."
)
},
)
max_target_length: Optional[int] = field(
default=128,
metadata={
"help": (
"The maximum total sequence length for target text after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded."
)
},
)
val_max_target_length: Optional[int] = field(
default=None,
metadata={
"help": (
"The maximum total sequence length for validation target text after tokenization. Sequences longer "
"than this will be truncated, sequences shorter will be padded. Will default to `max_target_length`."
"This argument is also used to override the ``max_length`` param of ``model.generate``, which is used "
"during ``evaluate`` and ``predict``."
)
},
)
pad_to_max_length: bool = field(
default=False,
metadata={
"help": (
"Whether to pad all samples to model maximum sentence length. "
"If False, will pad the samples dynamically when batching to the maximum length in the batch. More "
"efficient on GPU but very bad for TPU."
)
},
)
max_train_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of training examples to this "
"value if set."
)
},
)
max_eval_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of evaluation examples to this "
"value if set."
)
},
)
max_predict_samples: Optional[int] = field(
default=None,
metadata={
"help": (
"For debugging purposes or quicker training, truncate the number of prediction examples to this "
"value if set."
)
},
)
num_beams: Optional[int] = field(
default=None,
metadata={
"help": (
"Number of beams to use for evaluation. This argument will be passed to ``model.generate``, "
"which is used during ``evaluate`` and ``predict``."
)
},
)
ignore_pad_token_for_loss: bool = field(
default=True,
metadata={
"help": "Whether to ignore the tokens corresponding to padded labels in the loss computation or not."
},
)
source_prefix: Optional[str] = field(
default=None, metadata={"help": "A prefix to add before every source text (useful for T5 models)."}
)
forced_bos_token: Optional[str] = field(
default=None,
metadata={
"help": (
"The token to force as the first generated token after the :obj:`decoder_start_token_id`.Useful for"
" multilingual models like :doc:`mBART <../model_doc/mbart>` where the first generated token needs to"
" be the target language token.(Usually it is the target language token)"
)
},
)
def __post_init__(self):
if self.dataset_name is None and self.train_file is None and self.validation_file is None:
raise ValueError("Need either a dataset name or a training/validation file.")
elif self.source_lang is None or self.target_lang is None:
raise ValueError("Need to specify the source language and the target language.")
# accepting both json and jsonl file extensions, as
# many jsonlines files actually have a .json extension
valid_extensions = ["json", "jsonl"]
if self.train_file is not None:
extension = self.train_file.split(".")[-1]
assert extension in valid_extensions, "`train_file` should be a jsonlines file."
if self.validation_file is not None:
extension = self.validation_file.split(".")[-1]
assert extension in valid_extensions, "`validation_file` should be a jsonlines file."
if self.val_max_target_length is None:
self.val_max_target_length = self.max_target_length
def main():
# See all possible arguments in src/transformers/training_args.py
# or by passing the --help flag to this script.
# We now keep distinct sets of args, for a cleaner separation of concerns.
parser = HfArgumentParser((ModelArguments, DataTrainingArguments, Seq2SeqTrainingArguments))
if len(sys.argv) == 2 and sys.argv[1].endswith(".json"):
# If we pass only one argument to the script and it's the path to a json file,
# let's parse it to get our arguments.
model_args, data_args, training_args = parser.parse_json_file(json_file=os.path.abspath(sys.argv[1]))
else:
model_args, data_args, training_args = parser.parse_args_into_dataclasses()
# Sending telemetry. Tracking the example usage helps us better allocate resources to maintain them. The
# information sent is the one passed as arguments along with your Python/PyTorch versions.
send_example_telemetry("run_translation", model_args, data_args)
# Setup logging
logging.basicConfig(
format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
datefmt="%m/%d/%Y %H:%M:%S",
handlers=[logging.StreamHandler(sys.stdout)],
)
log_level = training_args.get_process_log_level()
logger.setLevel(log_level)
datasets.utils.logging.set_verbosity(log_level)
transformers.utils.logging.set_verbosity(log_level)
transformers.utils.logging.enable_default_handler()
transformers.utils.logging.enable_explicit_format()
# Log on each process the small summary:
logger.warning(
f"Process rank: {training_args.local_rank}, device: {training_args.device}, n_gpu: {training_args.n_gpu}"
+ f"distributed training: {bool(training_args.local_rank != -1)}, 16-bits training: {training_args.fp16}"
)
logger.info(f"Training/evaluation parameters {training_args}")
if data_args.source_prefix is None and model_args.model_name_or_path in [
"t5-small",
"t5-base",
"t5-large",
"t5-3b",
"t5-11b",
]:
logger.warning(
"You're running a t5 model but didn't provide a source prefix, which is expected, e.g. with "
"`--source_prefix 'translate English to German: ' `"
)
# Detecting last checkpoint.
last_checkpoint = None
if os.path.isdir(training_args.output_dir) and training_args.do_train and not training_args.overwrite_output_dir:
last_checkpoint = get_last_checkpoint(training_args.output_dir)
if last_checkpoint is None and len(os.listdir(training_args.output_dir)) > 0:
raise ValueError(
f"Output directory ({training_args.output_dir}) already exists and is not empty. "
"Use --overwrite_output_dir to overcome."
)
elif last_checkpoint is not None and training_args.resume_from_checkpoint is None:
logger.info(
f"Checkpoint detected, resuming training at {last_checkpoint}. To avoid this behavior, change "
"the `--output_dir` or add `--overwrite_output_dir` to train from scratch."
)
# Set seed before initializing model.
set_seed(training_args.seed)
# Get the datasets: you can either provide your own JSON training and evaluation files (see below)
# or just provide the name of one of the public datasets available on the hub at https://huggingface.co/datasets/
# (the dataset will be downloaded automatically from the datasets Hub).
#
# For translation, only JSON files are supported, with one field named "translation" containing two keys for the
# source and target languages (unless you adapt what follows).
#
# In distributed training, the load_dataset function guarantee that only one local process can concurrently
# download the dataset.
if data_args.dataset_name is not None:
# Downloading and loading a dataset from the hub.
raw_datasets = load_dataset(
data_args.dataset_name,
data_args.dataset_config_name,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
else:
data_files = {}
if data_args.train_file is not None:
data_files["train"] = data_args.train_file
extension = data_args.train_file.split(".")[-1]
if data_args.validation_file is not None:
data_files["validation"] = data_args.validation_file
extension = data_args.validation_file.split(".")[-1]
if data_args.test_file is not None:
data_files["test"] = data_args.test_file
extension = data_args.test_file.split(".")[-1]
raw_datasets = load_dataset(
extension,
data_files=data_files,
cache_dir=model_args.cache_dir,
use_auth_token=True if model_args.use_auth_token else None,
)
# See more about loading any type of standard or custom dataset (from files, python dict, pandas DataFrame, etc) at
# https://huggingface.co/docs/datasets/loading_datasets.html.
# Load pretrained model and tokenizer
#
# Distributed training:
# The .from_pretrained methods guarantee that only one local process can concurrently
# download model & vocab.
config = AutoConfig.from_pretrained(
model_args.config_name if model_args.config_name else model_args.model_name_or_path,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
tokenizer = AutoTokenizer.from_pretrained(
model_args.tokenizer_name if model_args.tokenizer_name else model_args.model_name_or_path,
cache_dir=model_args.cache_dir,
use_fast=model_args.use_fast_tokenizer,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
model = AutoModelForSeq2SeqLM.from_pretrained(
model_args.model_name_or_path,
from_tf=bool(".ckpt" in model_args.model_name_or_path),
config=config,
cache_dir=model_args.cache_dir,
revision=model_args.model_revision,
use_auth_token=True if model_args.use_auth_token else None,
)
if model_args.model_name_or_path.startswith('google/t5'):
for name, param in model.named_parameters():
if name.startswith('encoder.block'):
num_block = name.split('.')[2]
num_layer = name.split('.')[4]
if int(num_block) and int(num_layer) in [0,1]:
param.requires_grad = False
frozen_layers=[(name,param.requires_grad) for (name, param) in model.named_parameters() if not param.requires_grad]
print('\n\nFrozen layers:')
print(frozen_layers,'\n\n')
model.resize_token_embeddings(len(tokenizer))
# Set decoder_start_token_id
if model.config.decoder_start_token_id is None and isinstance(tokenizer, (MBartTokenizer, MBartTokenizerFast)):
if isinstance(tokenizer, MBartTokenizer):
model.config.decoder_start_token_id = tokenizer.lang_code_to_id[data_args.target_lang]
else:
model.config.decoder_start_token_id = tokenizer.convert_tokens_to_ids(data_args.target_lang)
if model.config.decoder_start_token_id is None:
raise ValueError("Make sure that `config.decoder_start_token_id` is correctly defined")
prefix = data_args.source_prefix if data_args.source_prefix is not None else ""
if 'classification' not in prefix:
raise RuntimeError('Not found "classification" prefix!')
prefix = prefix.strip()
if not prefix.endswith(':'):
prefix += ':'
prefix += ' '
logger.info(f'Using translation prefix: "{prefix}"')
# Preprocessing the datasets.
# We need to tokenize inputs and targets.
if training_args.do_train:
column_names = raw_datasets["train"].column_names
elif training_args.do_eval:
column_names = raw_datasets["validation"].column_names
elif training_args.do_predict:
column_names = raw_datasets["test"].column_names
else:
logger.info("There is nothing to do. Please pass `do_train`, `do_eval` and/or `do_predict`.")
return
# For translation we set the codes of our source and target languages (only useful for mBART, the others will
# ignore those attributes).
if isinstance(tokenizer, tuple(MULTILINGUAL_TOKENIZERS)):
assert data_args.target_lang is not None and data_args.source_lang is not None, (
f"{tokenizer.__class__.__name__} is a multilingual tokenizer which requires --source_lang and "
"--target_lang arguments."
)
tokenizer.src_lang = 'text'
tokenizer.tgt_lang = 'label'
# For multilingual translation models like mBART-50 and M2M100 we need to force the target language token
# as the first generated token. We ask the user to explicitly provide this as --forced_bos_token argument.
forced_bos_token_id = (
tokenizer.lang_code_to_id[data_args.forced_bos_token] if data_args.forced_bos_token is not None else None
)
model.config.forced_bos_token_id = forced_bos_token_id
# Get the language codes for input/target.
source_lang = 'text'
target_lang = 'label'
# Temporarily set max_target_length for training.
max_target_length = data_args.max_target_length
padding = "max_length" if data_args.pad_to_max_length else False
if training_args.label_smoothing_factor > 0 and not hasattr(model, "prepare_decoder_input_ids_from_labels"):
logger.warning(
"label_smoothing is enabled but the `prepare_decoder_input_ids_from_labels` method is not defined for"
f"`{model.__class__.__name__}`. This will lead to loss being calculated twice and will take up more memory"
)
def preprocess_function(examples):
inputs = [ex for ex in examples[source_lang]]
targets = [ex for ex in examples[target_lang]]
inputs = [prefix + inp for inp in inputs]
model_inputs = tokenizer(inputs, max_length=data_args.max_source_length, padding=padding, truncation=True)
# Tokenize targets with the `text_target` keyword argument
labels = tokenizer(text_target=targets, max_length=max_target_length, padding=padding, truncation=True)
# If we are padding here, replace all tokenizer.pad_token_id in the labels by -100 when we want to ignore
# padding in the loss.
if padding == "max_length" and data_args.ignore_pad_token_for_loss:
labels["input_ids"] = [
[(l if l != tokenizer.pad_token_id else -100) for l in label] for label in labels["input_ids"]
]
model_inputs["labels"] = labels["input_ids"]
return model_inputs
if training_args.do_train:
if "train" not in raw_datasets:
raise ValueError("--do_train requires a train dataset")
train_dataset = raw_datasets["train"]
if data_args.max_train_samples is not None:
max_train_samples = min(len(train_dataset), data_args.max_train_samples)
train_dataset = train_dataset.select(range(max_train_samples))
with training_args.main_process_first(desc="train dataset map pre-processing"):
train_dataset = train_dataset.map(
preprocess_function,
batched=True,
num_proc=data_args.preprocessing_num_workers,
remove_columns=column_names,
load_from_cache_file=not data_args.overwrite_cache,
desc="Running tokenizer on train dataset",
)
if training_args.do_eval:
max_target_length = data_args.val_max_target_length
if "validation" not in raw_datasets:
raise ValueError("--do_eval requires a validation dataset")
eval_dataset = raw_datasets["validation"]
if data_args.max_eval_samples is not None:
max_eval_samples = min(len(eval_dataset), data_args.max_eval_samples)
label_to_indexes = defaultdict(list)
for index, eval_sample in enumerate(eval_dataset):
label_to_indexes[eval_sample['label']].append(index)
max_samples_per_label = int(max_eval_samples / len(label_to_indexes))
eval_sample_indexes = []
for label, indexes in label_to_indexes.items():
eval_sample_indexes.extend(indexes[:max_samples_per_label])
logger.info(f"Set {max_samples_per_label} samples for {label}-class")
eval_sample_indexes.sort()
eval_dataset = eval_dataset.select(eval_sample_indexes)
with training_args.main_process_first(desc="validation dataset map pre-processing"):
eval_dataset = eval_dataset.map(
preprocess_function,
batched=True,
num_proc=data_args.preprocessing_num_workers,
remove_columns=column_names,
load_from_cache_file=not data_args.overwrite_cache,
desc="Running tokenizer on validation dataset",
)
if training_args.do_predict:
max_target_length = data_args.val_max_target_length
if "test" not in raw_datasets:
raise ValueError("--do_predict requires a test dataset")
predict_dataset = raw_datasets["test"]
if data_args.max_predict_samples is not None:
max_predict_samples = min(len(predict_dataset), data_args.max_predict_samples)
predict_dataset = predict_dataset.select(range(max_predict_samples))
with training_args.main_process_first(desc="prediction dataset map pre-processing"):
predict_dataset = predict_dataset.map(
preprocess_function,
batched=True,
num_proc=data_args.preprocessing_num_workers,
remove_columns=column_names,
load_from_cache_file=not data_args.overwrite_cache,
desc="Running tokenizer on prediction dataset",
)
# Data collator
label_pad_token_id = -100 if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id
if data_args.pad_to_max_length:
data_collator = default_data_collator
else:
data_collator = DataCollatorForSeq2Seq(
tokenizer,
model=model,
label_pad_token_id=label_pad_token_id,
pad_to_multiple_of=8 if training_args.fp16 else None,
)
# Metric
metric = evaluate.load("sacrebleu")
metric_accuracy = evaluate.load("accuracy")
def postprocess_text(preds, labels):
preds = [pred.strip() for pred in preds]
labels = [[label.strip()] for label in labels]
return preds, labels
def compute_metrics(eval_preds):
preds, labels = eval_preds
if isinstance(preds, tuple):
preds = preds[0]
decoded_preds = tokenizer.batch_decode(preds, skip_special_tokens=True)
if data_args.ignore_pad_token_for_loss:
# Replace -100 in the labels as we can't decode them.
labels = np.where(labels != -100, labels, tokenizer.pad_token_id)
decoded_labels = tokenizer.batch_decode(labels, skip_special_tokens=True)
# Some simple post-processing
decoded_preds, decoded_labels = postprocess_text(decoded_preds, decoded_labels)
decoded_preds_accuracy = [MAP_LABEL_TRANSLATION.get(decoded_pred, -1) for decoded_pred in decoded_preds]
decoded_labels_accuracy = [MAP_LABEL_TRANSLATION.get(decoded_label[0], -1) for decoded_label in decoded_labels]
result = metric.compute(predictions=decoded_preds, references=decoded_labels)
result_accuracy = metric_accuracy.compute(predictions=decoded_preds_accuracy, references=decoded_labels_accuracy)
result = {"bleu": result["score"], "accuracy": result_accuracy["accuracy"]}
prediction_lens = [np.count_nonzero(pred != tokenizer.pad_token_id) for pred in preds]
result["gen_len"] = np.mean(prediction_lens)
result = {k: round(v, 4) for k, v in result.items()}
return result
# Initialize our Trainer
trainer = Seq2SeqTrainer(
model=model,
args=training_args,
train_dataset=train_dataset if training_args.do_train else None,
eval_dataset=eval_dataset if training_args.do_eval else None,
tokenizer=tokenizer,
data_collator=data_collator,
compute_metrics=compute_metrics if training_args.predict_with_generate else None,
)
# Training
if training_args.do_train:
checkpoint = None
if training_args.resume_from_checkpoint is not None:
checkpoint = training_args.resume_from_checkpoint
elif last_checkpoint is not None:
checkpoint = last_checkpoint
train_result = trainer.train(resume_from_checkpoint=checkpoint)
trainer.save_model() # Saves the tokenizer too for easy upload
metrics = train_result.metrics
max_train_samples = (
data_args.max_train_samples if data_args.max_train_samples is not None else len(train_dataset)
)
metrics["train_samples"] = min(max_train_samples, len(train_dataset))
trainer.log_metrics("train", metrics)
trainer.save_metrics("train", metrics)
trainer.save_state()
# Evaluation
results = {}
max_length = (
training_args.generation_max_length
if training_args.generation_max_length is not None
else data_args.val_max_target_length
)
num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
if training_args.do_eval:
logger.info("*** Evaluate ***")
metrics = trainer.evaluate(max_length=max_length, num_beams=num_beams, metric_key_prefix="eval")
max_eval_samples = data_args.max_eval_samples if data_args.max_eval_samples is not None else len(eval_dataset)
metrics["eval_samples"] = min(max_eval_samples, len(eval_dataset))
trainer.log_metrics("eval", metrics)
trainer.save_metrics("eval", metrics)
if training_args.do_predict:
logger.info("*** Predict ***")
predict_results = trainer.predict(
predict_dataset, metric_key_prefix="predict", max_length=max_length, num_beams=num_beams
)
metrics = predict_results.metrics
max_predict_samples = (
data_args.max_predict_samples if data_args.max_predict_samples is not None else len(predict_dataset)
)
metrics["predict_samples"] = min(max_predict_samples, len(predict_dataset))
trainer.log_metrics("predict", metrics)
trainer.save_metrics("predict", metrics)
if trainer.is_world_process_zero():
if training_args.predict_with_generate:
predictions = tokenizer.batch_decode(
predict_results.predictions, skip_special_tokens=True, clean_up_tokenization_spaces=True
)
predictions = [pred.strip() for pred in predictions]
output_prediction_file = os.path.join(training_args.output_dir, "generated_predictions.txt")
with open(output_prediction_file, "w", encoding="utf-8") as writer:
writer.write("\n".join(predictions))
kwargs = {"finetuned_from": model_args.model_name_or_path, "tasks": "translation"}
if data_args.dataset_name is not None:
kwargs["dataset_tags"] = data_args.dataset_name
if data_args.dataset_config_name is not None:
kwargs["dataset_args"] = data_args.dataset_config_name
kwargs["dataset"] = f"{data_args.dataset_name} {data_args.dataset_config_name}"
else:
kwargs["dataset"] = data_args.dataset_name
languages = [l for l in [data_args.source_lang, data_args.target_lang] if l is not None]
if len(languages) > 0:
kwargs["language"] = languages
if training_args.push_to_hub:
trainer.push_to_hub(**kwargs)
else:
trainer.create_model_card(**kwargs)
return results
def _mp_fn(index):
# For xla_spawn (TPUs)
main()
if __name__ == "__main__":
main()