Create classifier.py
Browse files- classifier.py +27 -0
classifier.py
ADDED
|
@@ -0,0 +1,27 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# classifier.py
|
| 2 |
+
import torch
|
| 3 |
+
from model_loader import model, tokenizer
|
| 4 |
+
|
| 5 |
+
def classify_toxic_comment(comment):
|
| 6 |
+
"""
|
| 7 |
+
Classify a comment as toxic or non-toxic using the fine-tuned XLM-RoBERTa model.
|
| 8 |
+
Returns the prediction label, confidence, and color for UI display.
|
| 9 |
+
"""
|
| 10 |
+
if not comment.strip():
|
| 11 |
+
return "Error: Please enter a comment.", None, None
|
| 12 |
+
|
| 13 |
+
# Tokenize the input comment
|
| 14 |
+
inputs = tokenizer(comment, return_tensors="pt", truncation=True, padding=True, max_length=512)
|
| 15 |
+
|
| 16 |
+
# Run inference
|
| 17 |
+
with torch.no_grad():
|
| 18 |
+
outputs = model(**inputs)
|
| 19 |
+
logits = outputs.logits
|
| 20 |
+
|
| 21 |
+
# Get the predicted class (0 = non-toxic, 1 = toxic)
|
| 22 |
+
predicted_class = torch.argmax(logits, dim=1).item()
|
| 23 |
+
label = "Toxic" if predicted_class == 1 else "Non-Toxic"
|
| 24 |
+
confidence = torch.softmax(logits, dim=1)[0][predicted_class].item()
|
| 25 |
+
label_color = "red" if label == "Toxic" else "green"
|
| 26 |
+
|
| 27 |
+
return f"Prediction: {label}", confidence, label_color
|