alihan/textattack-nlp-transformer
1
0
Fork 0
mirror of https://github.com/QData/TextAttack.git synced 2021-10-13 00:05:06 +03:00
Code Issues Projects Releases Wiki Activity

fix model

Browse Source
This commit is contained in:
uvafan
2020-03-05 22:29:07 -05:00
parent bc5e4d57a7
commit 0b6da2a475
4 changed files with 10 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
scripts/run_attack_single_threaded.py
View File

@@ -45,12 +45,12 @@ def run(args):
if not text:
continue
tokenized_text = textattack.shared.tokenized_text.TokenizedText(text, model.tokenizer)
tokenized_text = textattack.shared.tokenized_text.TokenizedText(text, goal_function.model.tokenizer)
result = goal_function.get_results([tokenized_text])[0]
result = goal_function.get_results([tokenized_text], goal_function.get_output(tokenized_text))[0]
print('Attacking...')
result = next(attack.attack_dataset([(result.output, text, False)]))
result = next(attack.attack_dataset([(result.output, text)]))
print(result.__str__(color_method='stdout'))
else:

4
textattack/goal_functions/goal_function.py
View File

@@ -14,12 +14,16 @@ class GoalFunction:
"""
def __init__(self, model):
self.model = model
self.num_queries = 0
self._call_model_cache = lru.LRU(2**18)
def should_skip(self, tokenized_text, correct_output):
model_outputs = self._call_model([tokenized_text])
return self._is_goal_complete(model_outputs[0], correct_output)
def get_output(self, tokenized_text):
return self._get_displayed_output(self._call_model([tokenized_text])[0])
def get_results(self, tokenized_text_list, correct_output):
"""
For each tokenized_text object in tokenized_text_list, returns a result consisting of whether or not the goal has been achieved, the output for display purposes, and a score.

2
textattack/models/helpers/bert_for_classification.py
View File

@@ -17,7 +17,7 @@ class BERTForClassification:
"""
def __init__(self, model_path, num_labels=2, entailment=False):
#model_file_path = utils.download_if_needed(model_path)
model_file_path = '/p/qdata/jm8wx/research/text_attacks/RobustNLP/BertClassifier/outputs-counterfit/mr-uncased-2020-03-04-23:05'
model_file_path = utils.download_if_needed(model_path)
self.model = BertForSequenceClassification.from_pretrained(
model_file_path, num_labels=num_labels)
self.model.to(utils.get_device())

2
textattack/tokenizers/bert_tokenizer.py
View File

@@ -6,7 +6,7 @@ class BERTTokenizer(Tokenizer):
any type of tokenization, be it word, wordpiece, or character-based.
"""
def __init__(self, model_path='bert-base-uncased', max_seq_length=256):
self.tokenizer = BertTokenizer.from_pretrained('bert-base-uncased')
self.tokenizer = BertTokenizer.from_pretrained(model_path)
self.max_seq_length = max_seq_length
def convert_text_to_tokens(self, input_text):