Upload 2 files
Browse files- moderation_model.pth +3 -0
- test.py +71 -0
moderation_model.pth
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:cb5f1cbf7c576b2d1ea0ee801d90178b2d392ba37256573bc8454c38ae521854
|
3 |
+
size 204952
|
test.py
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
import torch
|
4 |
+
import torch.nn as nn
|
5 |
+
import torch.optim as optim
|
6 |
+
import torch.nn.functional as F
|
7 |
+
from torch.utils.data import Dataset, DataLoader
|
8 |
+
|
9 |
+
from transformers import AutoTokenizer, AutoModel
|
10 |
+
|
11 |
+
device = "cuda" if torch.cuda.is_available() else "cpu"
|
12 |
+
|
13 |
+
tokenizer_embeddings = AutoTokenizer.from_pretrained('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2')
|
14 |
+
model_embeddings = AutoModel.from_pretrained('sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2').to(device)
|
15 |
+
|
16 |
+
class ModerationModel(nn.Module):
|
17 |
+
def __init__(self):
|
18 |
+
input_size = 384
|
19 |
+
hidden_size = 128
|
20 |
+
output_size = 11
|
21 |
+
super(ModerationModel, self).__init__()
|
22 |
+
self.fc1 = nn.Linear(input_size, hidden_size)
|
23 |
+
self.fc2 = nn.Linear(hidden_size, output_size)
|
24 |
+
|
25 |
+
def forward(self, x):
|
26 |
+
x = F.relu(self.fc1(x))
|
27 |
+
x = self.fc2(x)
|
28 |
+
return x
|
29 |
+
|
30 |
+
def mean_pooling(model_output, attention_mask):
|
31 |
+
token_embeddings = model_output[0]
|
32 |
+
input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
|
33 |
+
return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)
|
34 |
+
|
35 |
+
def getEmbeddings(sentences):
|
36 |
+
encoded_input = tokenizer_embeddings(sentences, padding=True, truncation=True, return_tensors='pt').to(device)
|
37 |
+
with torch.no_grad():
|
38 |
+
model_output = model_embeddings(**encoded_input)
|
39 |
+
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
|
40 |
+
return sentence_embeddings.cpu()
|
41 |
+
|
42 |
+
def getEmb(text):
|
43 |
+
sentences = [text]
|
44 |
+
sentence_embeddings = getEmbeddings(sentences)
|
45 |
+
return sentence_embeddings.tolist()[0]
|
46 |
+
|
47 |
+
def predict(model, embeddings):
|
48 |
+
model.eval()
|
49 |
+
with torch.no_grad():
|
50 |
+
embeddings_tensor = torch.tensor(embeddings, dtype=torch.float)
|
51 |
+
outputs = model(embeddings_tensor.unsqueeze(0))
|
52 |
+
predicted_scores = torch.sigmoid(outputs)
|
53 |
+
predicted_scores = predicted_scores.squeeze(0).tolist()
|
54 |
+
category_names = ["harassment", "harassment-threatening", "hate", "hate-threatening", "self-harm", "self-harm-instructions", "self-harm-intent", "sexual", "sexual-minors", "violence", "violence-graphic"]
|
55 |
+
|
56 |
+
result = {category: score for category, score in zip(category_names, predicted_scores)}
|
57 |
+
detected = {category: score > 0.5 for category, score in zip(category_names, predicted_scores)}
|
58 |
+
detect_value = any(detected.values())
|
59 |
+
|
60 |
+
return {"category_scores": result, 'detect': detected, 'detected': detect_value}
|
61 |
+
|
62 |
+
|
63 |
+
print('Load model')
|
64 |
+
moderation = ModerationModel()
|
65 |
+
moderation.load_state_dict(torch.load('moderation_model.pth'))
|
66 |
+
|
67 |
+
text = "I want to kill them."
|
68 |
+
|
69 |
+
embeddings_for_prediction = getEmb(text)
|
70 |
+
prediction = predict(moderation, embeddings_for_prediction)
|
71 |
+
print(json.dumps(prediction,indent=4))
|