Newer
Older
import torch
from transformers import AutoTokenizer, AutoModelForSequenceClassification
### from models.py
from pretrained.models import *
import json
from nltk.tokenize.treebank import TreebankWordDetokenizer
device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = AutoTokenizer.from_pretrained("Hate-speech-CNERG/bert-base-uncased-hatexplain-rationale-two")
model = \
Model_Rational_Label.from_pretrained(
"Hate-speech-CNERG/bert-base-uncased-hatexplain-rationale-two"
)
model = model.to(device)
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
# inputs = tokenizer(['He is a great guy', 'He is a greeeeeet guy'],
# return_tensors="pt", padding=True).to(device)
# print(inputs['input_ids'])
# prediction_logits, _ = model(input_ids=inputs['input_ids'],attention_mask=inputs['attention_mask'])
# softmax = torch.nn.Softmax(dim=1)
# probs = softmax(prediction_logits)
# print(f"Normal: {probs[0][0]}\nHatespeech: {probs[0][1]}")
# print(f"Normal: {probs[1][0]}\nHatespeech: {probs[1][1]}")
# Load test dataset
with open('data/post_id_divisions.json') as splits:
data = json.load(splits)
test_ids = data['test']
def dataset(ids):
with open('data/dataset.json') as data_file:
data = json.load(data_file)
for i in ids:
yield data[i]
counter = 0
batchsize = 8
for post in dataset(test_ids):
# if counter == 5:
# break
# counter += 1
detokenized = TreebankWordDetokenizer().detokenize(post["post_tokens"])
# ATTACK HERE
inputs = tokenizer(detokenized, return_tensors="pt", padding=True).to(device)
prediction_logits, _ = model(input_ids=inputs['input_ids'],attention_mask=inputs['attention_mask'])
softmax = torch.nn.Softmax(dim=1)
probs = softmax(prediction_logits)
print(f"Normal: {probs[0][0]}\nHatespeech: {probs[0][1]}\n\n")
print(f"Normal: {probs[1][0]}\nHatespeech: {probs[1][1]}\n\n")
break
# print("------------------")
# print(post["post_id"])
# print(post["annotators"][0]["label"])
# print(TreebankWordDetokenizer().detokenize(post["post_tokens"]))