Spaces:
Sleeping
Sleeping
hk-bt-rnd
commited on
Commit
•
ad5ee12
1
Parent(s):
c5550af
Init space
Browse files- __pycache__/model.cpython-310.pyc +0 -0
- app.py +87 -0
- model.py +37 -0
- requirements.txt +76 -0
- weight.pt +3 -0
__pycache__/model.cpython-310.pyc
ADDED
Binary file (1.24 kB). View file
|
|
app.py
ADDED
@@ -0,0 +1,87 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
import numpy as np
|
3 |
+
from PIL import Image
|
4 |
+
from matplotlib import cm
|
5 |
+
import torch
|
6 |
+
from transformers import AutoTokenizer, AutoModel, AutoConfig
|
7 |
+
from model import Classifier
|
8 |
+
import torch.nn as nn
|
9 |
+
import torch.nn.functional as F
|
10 |
+
|
11 |
+
# Load model directly
|
12 |
+
MODEL_NAME = "cahya/roberta-base-indonesian-522M"
|
13 |
+
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
|
14 |
+
class_names = ['Action', 'Adventure', 'Comedy', 'Drama', 'Fantasy', 'Romance', 'Sci-Fi']
|
15 |
+
config = AutoConfig.from_pretrained(MODEL_NAME)
|
16 |
+
transformer = AutoModel.from_pretrained(MODEL_NAME)
|
17 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
|
18 |
+
cp = torch.load(r"weight.pt", map_location="cpu")
|
19 |
+
transformer.load_state_dict(cp['w_t'])
|
20 |
+
classifier = Classifier(input_size = config.hidden_size, output_sizes = [1, 1, 1, 3, 5])
|
21 |
+
classifier.load_state_dict(cp['w_c'])
|
22 |
+
|
23 |
+
transformer.to(device)
|
24 |
+
classifier.to(device)
|
25 |
+
|
26 |
+
target_names = ["Individual", 'Group']
|
27 |
+
strength_names = ["Weak", 'Moderate', 'Strong']
|
28 |
+
type_names = ['Religion','Race','Physical','Gender','Other']
|
29 |
+
|
30 |
+
act_sig = nn.Sigmoid()
|
31 |
+
act_soft = nn.Softmax()
|
32 |
+
|
33 |
+
def predict(sentence):
|
34 |
+
# Tokenize the input sentence
|
35 |
+
inputs = tokenizer(sentence,
|
36 |
+
add_special_tokens = True, \
|
37 |
+
max_length = 256, \
|
38 |
+
padding = "max_length", \
|
39 |
+
truncation = True,
|
40 |
+
return_tensors='pt')
|
41 |
+
|
42 |
+
input_ids = inputs['input_ids'].to(device)
|
43 |
+
att_masks = inputs['attention_mask'].to(device)
|
44 |
+
|
45 |
+
# Get model predictions
|
46 |
+
with torch.no_grad():
|
47 |
+
out = transformer(input_ids, attention_mask=att_masks)
|
48 |
+
logits = out.pooler_output
|
49 |
+
out = classifier(logits)
|
50 |
+
hs_out, abusive_out, target_out, strength_out, type_out = out[0], out[1], out[2], out[3], out[4]
|
51 |
+
hs_act, abusive_act, target_act, strength_act, type_act = act_sig(hs_out).squeeze(), \
|
52 |
+
act_sig(abusive_out).squeeze(), act_sig(target_out).squeeze(0), act_soft(strength_out), act_sig(type_out).squeeze(0)
|
53 |
+
|
54 |
+
# Interpret the predictions
|
55 |
+
is_hate_speech = bool(hs_act >= 0.5)
|
56 |
+
is_abusive = bool(abusive_act >= 0.5)
|
57 |
+
hate_speech_target = int(target_act >= 0.5)
|
58 |
+
hate_speech_strength = strength_act.argmax().item()
|
59 |
+
if is_hate_speech:
|
60 |
+
hate_speech_target_label = target_names[hate_speech_target]
|
61 |
+
hate_speech_strength_label = strength_names[hate_speech_strength]
|
62 |
+
hate_speech_type_label = []
|
63 |
+
print('target', target_act)
|
64 |
+
print('strength', strength_act)
|
65 |
+
|
66 |
+
for idx, prob in enumerate(type_act):
|
67 |
+
if prob >= 0.5:
|
68 |
+
hate_speech_type_label.append(type_names[idx])
|
69 |
+
if len(hate_speech_type_label) == 0:
|
70 |
+
hate_speech_type_label.append("Other")
|
71 |
+
else:
|
72 |
+
hate_speech_target_label = "Non-HS"
|
73 |
+
hate_speech_strength_label = "Non-HS"
|
74 |
+
hate_speech_type_label = "Non-HS"
|
75 |
+
|
76 |
+
return is_hate_speech, is_abusive, hate_speech_target_label, hate_speech_strength_label, {"hs_type":hate_speech_type_label}
|
77 |
+
|
78 |
+
# Create the Gradio interface
|
79 |
+
iface = gr.Interface(fn=predict, inputs=gr.Textbox(label="Enter a sentence"), outputs=[
|
80 |
+
gr.Label(label="Is Hate Speech"),
|
81 |
+
gr.Label(label="Is Abusive"),
|
82 |
+
gr.Label(label="Hate Speech Target"),
|
83 |
+
gr.Label(label="Hate Speech Strength"),
|
84 |
+
gr.JSON(label="Hate Speech Type")
|
85 |
+
], title="Hate Speech Detection")
|
86 |
+
iface.launch() # Launches the mini app!
|
87 |
+
|
model.py
ADDED
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch.nn as nn
|
2 |
+
import torchvision.models as models
|
3 |
+
import torch
|
4 |
+
from transformers import AutoTokenizer, AutoModel, AutoConfig
|
5 |
+
|
6 |
+
class Classifier(nn.Module):
|
7 |
+
def __init__(self, input_size = 512, output_sizes = [1], dropout_rate = 0.1):
|
8 |
+
super(Classifier, self).__init__()
|
9 |
+
|
10 |
+
self.hs_head = nn.Sequential(
|
11 |
+
nn.Dropout(dropout_rate),
|
12 |
+
nn.Linear(input_size, output_sizes[0])
|
13 |
+
)
|
14 |
+
|
15 |
+
self.abusive_head = nn.Sequential(
|
16 |
+
nn.Dropout(dropout_rate),
|
17 |
+
nn.Linear(input_size, output_sizes[1])
|
18 |
+
)
|
19 |
+
|
20 |
+
self.target_head = nn.Sequential(
|
21 |
+
nn.Dropout(dropout_rate),
|
22 |
+
nn.Linear(input_size, output_sizes[2])
|
23 |
+
)
|
24 |
+
|
25 |
+
self.strength_head = nn.Sequential(
|
26 |
+
nn.Dropout(dropout_rate),
|
27 |
+
nn.Linear(input_size, output_sizes[3])
|
28 |
+
)
|
29 |
+
|
30 |
+
self.type_head = nn.Sequential(
|
31 |
+
nn.Dropout(dropout_rate),
|
32 |
+
nn.Linear(input_size, output_sizes[4])
|
33 |
+
)
|
34 |
+
|
35 |
+
def forward(self, input):
|
36 |
+
return self.hs_head(input), self.abusive_head(input), self.target_head(input), \
|
37 |
+
self.strength_head(input), self.type_head(input)
|
requirements.txt
ADDED
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
aiofiles==23.2.1
|
2 |
+
altair==5.2.0
|
3 |
+
annotated-types==0.6.0
|
4 |
+
anyio==4.3.0
|
5 |
+
attrs==23.2.0
|
6 |
+
certifi==2024.2.2
|
7 |
+
charset-normalizer==3.3.2
|
8 |
+
click==8.1.7
|
9 |
+
colorama==0.4.6
|
10 |
+
contourpy==1.2.0
|
11 |
+
cycler==0.12.1
|
12 |
+
exceptiongroup==1.2.0
|
13 |
+
fastapi==0.110.0
|
14 |
+
ffmpy==0.3.2
|
15 |
+
filelock==3.13.1
|
16 |
+
fonttools==4.50.0
|
17 |
+
fsspec==2024.3.1
|
18 |
+
gradio==4.22.0
|
19 |
+
gradio_client==0.13.0
|
20 |
+
h11==0.14.0
|
21 |
+
httpcore==1.0.4
|
22 |
+
httpx==0.27.0
|
23 |
+
huggingface-hub==0.21.4
|
24 |
+
idna==3.6
|
25 |
+
importlib_resources==6.3.2
|
26 |
+
Jinja2==3.1.3
|
27 |
+
jsonschema==4.21.1
|
28 |
+
jsonschema-specifications==2023.12.1
|
29 |
+
kiwisolver==1.4.5
|
30 |
+
markdown-it-py==3.0.0
|
31 |
+
MarkupSafe==2.1.5
|
32 |
+
matplotlib==3.8.3
|
33 |
+
mdurl==0.1.2
|
34 |
+
mpmath==1.3.0
|
35 |
+
networkx==3.2.1
|
36 |
+
numpy==1.26.4
|
37 |
+
orjson==3.9.15
|
38 |
+
packaging==24.0
|
39 |
+
pandas==2.2.1
|
40 |
+
pillow==10.2.0
|
41 |
+
pydantic==2.6.4
|
42 |
+
pydantic_core==2.16.3
|
43 |
+
pydub==0.25.1
|
44 |
+
Pygments==2.17.2
|
45 |
+
pyparsing==3.1.2
|
46 |
+
python-dateutil==2.9.0.post0
|
47 |
+
python-multipart==0.0.9
|
48 |
+
pytz==2024.1
|
49 |
+
PyYAML==6.0.1
|
50 |
+
referencing==0.34.0
|
51 |
+
regex==2023.12.25
|
52 |
+
requests==2.31.0
|
53 |
+
rich==13.7.1
|
54 |
+
rpds-py==0.18.0
|
55 |
+
ruff==0.3.3
|
56 |
+
safetensors==0.4.2
|
57 |
+
semantic-version==2.10.0
|
58 |
+
shellingham==1.5.4
|
59 |
+
six==1.16.0
|
60 |
+
sniffio==1.3.1
|
61 |
+
starlette==0.36.3
|
62 |
+
sympy==1.12
|
63 |
+
tokenizers==0.15.2
|
64 |
+
tomlkit==0.12.0
|
65 |
+
toolz==0.12.1
|
66 |
+
torch==2.2.1
|
67 |
+
torchaudio==2.2.1
|
68 |
+
torchvision==0.17.1
|
69 |
+
tqdm==4.66.2
|
70 |
+
transformers==4.38.2
|
71 |
+
typer==0.9.0
|
72 |
+
typing_extensions==4.10.0
|
73 |
+
tzdata==2024.1
|
74 |
+
urllib3==2.2.1
|
75 |
+
uvicorn==0.29.0
|
76 |
+
websockets==11.0.3
|
weight.pt
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:eb083732f7dd150113bba50f7f5125a4d3b83adf98db912d64394d38d9290e1b
|
3 |
+
size 504022203
|