Create app.py

app.py

@@ -0,0 +1,90 @@
+import gradio as gr
+import torch
+from transformers import AutoModel, AutoTokenizer
+
+# 1. Model Definition (Must Match Training Architecture)
+class BERTFakeNewsClassifier(torch.nn.Module):
+    def __init__(self, base_model):
+        super().__init__()
+        self.bert = base_model
+        self.classifier = torch.nn.Sequential(
+            torch.nn.Dropout(0.1),
+            torch.nn.Linear(768, 512),
+            torch.nn.ReLU(),
+            torch.nn.Linear(512, 2),
+            torch.nn.LogSoftmax(dim=1)
+        )
+    
+    def forward(self, input_ids, attention_mask):
+        outputs = self.bert(input_ids, attention_mask=attention_mask)
+        pooled = outputs.last_hidden_state[:, 0, 🙂  # Use [CLS] token
+        return self.classifier(pooled)
+
+# 2. Load Model (Optimized for Inference)
+def load_model_for_inference(model_path="KenLumod/ML-Project-Fake-Real-News-Detector-Final"):
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    
+    # Load components
+    tokenizer = AutoTokenizer.from_pretrained(model_path)
+    base_model = AutoModel.from_pretrained("bert-base-uncased")
+    model = BERTFakeNewsClassifier(base_model)
+    
+    # Load trained weights
+    state_dict = torch.load(f"{model_path}/pytorch_model.bin", map_location=device)
+    model.load_state_dict(state_dict)
+    model.to(device).eval()
+    
+    return model, tokenizer, device
+
+# 3. Prediction Function (Gradio-Compatible)
+def predict(text, model, tokenizer, device, max_length=128):
+    inputs = tokenizer(
+        text,
+        max_length=max_length,
+        padding="max_length",
+        truncation=True,
+        return_tensors="pt"
+    ).to(device)
+    
+    with torch.no_grad():
+        logits = model(**inputs)
+        probs = torch.exp(logits).cpu().numpy()[0]  # Convert log-probs to probabilities
+    
+    return {"REAL": float(probs[0]), "FAKE": float(probs[1])}
+
+# 4. Gradio Interface Builder
+def create_gradio_interface():
+    model, tokenizer, device = load_model_for_inference()
+    
+    def classify_text(text):
+        return predict(text, model, tokenizer, device)
+    
+    return gr.Interface(
+        fn=classify_text,
+        inputs=gr.Textbox(
+            label="News Content",
+            placeholder="Paste news article or headline here...",
+            lines=3
+        ),
+        outputs=gr.Label(
+            label="Detection Result",
+            num_top_classes=2
+        ),
+        examples=[
+            ["BREAKING: Trump arrested at Mar-a-Lago - Secret Service confirms"],
+            ["Congress passes bipartisan infrastructure bill after months of negotiation"],
+            ["NASA discovers alien city on Mars - photos leaked"]
+        ],
+        title="Fake News Detector (BERT)",
+        description="Classifies news content as REAL or FAKE using fine-tuned BERT",
+        allow_flagging="never"
+    )
+
+# 5. Launch
+if _name_ == "_main_":
+    demo = create_gradio_interface()
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=False  # Set to True for temporary public link
+    )