Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,13 +1,87 @@
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
-
def greet(name):
|
5 |
-
result = "Label Probabilities:\n" + f"African American: {str(round(0.797795832157135,2)*100)}\n"
|
6 |
-
+ f"Asian: {str(round(0.17413224279880524,2)*100)}\n"+ f"Latin: {str(round(0.0132269160822033,2)*100)}\n"+ f"White: {str(round(0.14844958670437336,2)*100)}"
|
7 |
-
|
8 |
-
return result
|
9 |
|
|
|
10 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
11 |
|
12 |
-
demo = gr.Interface(fn=greet, inputs="text", outputs="text")
|
13 |
demo.launch()
|
|
|
|
1 |
+
import torch
|
2 |
+
from torch import nn
|
3 |
+
from transformers import AutoModel, AutoTokenizer
|
4 |
import gradio as gr
|
5 |
|
6 |
+
# Check if CUDA is available
|
7 |
+
if torch.cuda.is_available():
|
8 |
+
device = torch.device("cuda")
|
9 |
+
else:
|
10 |
+
device = torch.device("cpu")
|
11 |
|
|
|
|
|
|
|
|
|
|
|
12 |
|
13 |
+
class RaceClassifier(nn.Module):
|
14 |
|
15 |
+
def __init__(self, n_classes):
|
16 |
+
super(RaceClassifier, self).__init__()
|
17 |
+
self.bert = AutoModel.from_pretrained("vinai/bertweet-base")
|
18 |
+
self.drop = nn.Dropout(p=0.3) # can be changed in future
|
19 |
+
self.out = nn.Linear(self.bert.config.hidden_size,
|
20 |
+
n_classes) # linear layer for the output with the number of classes
|
21 |
+
|
22 |
+
def forward(self, input_ids, attention_mask):
|
23 |
+
bert_output = self.bert(
|
24 |
+
input_ids=input_ids,
|
25 |
+
attention_mask=attention_mask
|
26 |
+
)
|
27 |
+
last_hidden_state = bert_output[0]
|
28 |
+
pooled_output = last_hidden_state[:, 0]
|
29 |
+
output = self.drop(pooled_output)
|
30 |
+
return self.out(output)
|
31 |
+
|
32 |
+
|
33 |
+
labels = {
|
34 |
+
0: "African American",
|
35 |
+
1: "Asian",
|
36 |
+
2: "Latin",
|
37 |
+
3: "White"
|
38 |
+
}
|
39 |
+
model_race = RaceClassifier(n_classes=4)
|
40 |
+
model_race.to(device)
|
41 |
+
model_race.load_state_dict(torch.load('best_model_race.pt'))
|
42 |
+
|
43 |
+
|
44 |
+
def predict(text):
|
45 |
+
sentences = [
|
46 |
+
text
|
47 |
+
]
|
48 |
+
|
49 |
+
tokenizer = AutoTokenizer.from_pretrained("vinai/bertweet-base", normalization=True)
|
50 |
+
|
51 |
+
encoded_sentences = tokenizer(
|
52 |
+
sentences,
|
53 |
+
padding=True,
|
54 |
+
truncation=True,
|
55 |
+
return_tensors='pt',
|
56 |
+
max_length=128,
|
57 |
+
)
|
58 |
+
|
59 |
+
input_ids = encoded_sentences["input_ids"].to(device)
|
60 |
+
attention_mask = encoded_sentences["attention_mask"].to(device)
|
61 |
+
|
62 |
+
model_race.eval()
|
63 |
+
with torch.no_grad():
|
64 |
+
outputs = model_race(input_ids, attention_mask)
|
65 |
+
probs = torch.nn.functional.softmax(outputs, dim=1)
|
66 |
+
predictions = torch.argmax(outputs, dim=1)
|
67 |
+
predictions = predictions.cpu().numpy()
|
68 |
+
|
69 |
+
output_string = ""
|
70 |
+
for i, prob in enumerate(probs[0]):
|
71 |
+
print(f"{labels[i]}: %{round(prob.item() * 100, 2)}")
|
72 |
+
output_string += f"{labels[i]}: %{round(prob.item() * 100, 2)}\n"
|
73 |
+
|
74 |
+
print(labels[predictions[0]])
|
75 |
+
output_string += f"Predicted as: {labels[predictions[0]]}"
|
76 |
+
|
77 |
+
return output_string
|
78 |
+
|
79 |
+
|
80 |
+
demo = gr.Interface(
|
81 |
+
fn=predict,
|
82 |
+
inputs=["text"],
|
83 |
+
outputs=["text"],
|
84 |
+
)
|
85 |
|
|
|
86 |
demo.launch()
|
87 |
+
|