SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[HANS] Fix label_list for RoBERTa/BART (class flipping) (#5196)

Browse Source 
* fix weirdness in roberta/bart for mnli trained checkpoints

* black compliance

* isort code check
This commit is contained in:
Victor SANH
2020-06-24 14:38:15 -04:00
committed by GitHub
parent fc24a93e64
commit 4965aee064
2 changed files with 41 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

40
examples/adversarial/utils_hans.py
View File

@@ -22,7 +22,17 @@ from typing import List, Optional, Union
import tqdm
from filelock import FileLock
from transformers import DataProcessor, PreTrainedTokenizer, is_tf_available, is_torch_available
from transformers import (
BartTokenizer,
BartTokenizerFast,
DataProcessor,
PreTrainedTokenizer,
RobertaTokenizer,
RobertaTokenizerFast,
XLMRobertaTokenizer,
is_tf_available,
is_torch_available,
)
logger = logging.getLogger(__name__)
@@ -105,6 +115,17 @@ if is_torch_available():
"dev" if evaluate else "train", tokenizer.__class__.__name__, str(max_seq_length), task,
),
)
label_list = processor.get_labels()
if tokenizer.__class__ in (
RobertaTokenizer,
RobertaTokenizerFast,
XLMRobertaTokenizer,
BartTokenizer,
BartTokenizerFast,
):
# HACK(label indices are swapped in RoBERTa pretrained model)
label_list[1], label_list[2] = label_list[2], label_list[1]
self.label_list = label_list
# Make sure only the first process in distributed training processes the dataset,
# and the others will use the cache.
@@ -116,7 +137,6 @@ if is_torch_available():
self.features = torch.load(cached_features_file)
else:
logger.info(f"Creating features from dataset file at {data_dir}")
label_list = processor.get_labels()
examples = (
processor.get_dev_examples(data_dir) if evaluate else processor.get_train_examples(data_dir)
@@ -133,6 +153,9 @@ if is_torch_available():
def __getitem__(self, i) -> InputFeatures:
return self.features[i]
def get_labels(self):
return self.label_list
if is_tf_available():
import tensorflow as tf
@@ -156,6 +179,16 @@ if is_tf_available():
):
processor = hans_processors[task]()
label_list = processor.get_labels()
if tokenizer.__class__ in (
RobertaTokenizer,
RobertaTokenizerFast,
XLMRobertaTokenizer,
BartTokenizer,
BartTokenizerFast,
):
# HACK(label indices are swapped in RoBERTa pretrained model)
label_list[1], label_list[2] = label_list[2], label_list[1]
self.label_list = label_list
examples = processor.get_dev_examples(data_dir) if evaluate else processor.get_train_examples(data_dir)
self.features = hans_convert_examples_to_features(examples, label_list, max_seq_length, tokenizer)
@@ -206,6 +239,9 @@ if is_tf_available():
def __getitem__(self, i) -> InputFeatures:
return self.features[i]
def get_labels(self):
return self.label_list
class HansProcessor(DataProcessor):
"""Processor for the HANS data set."""