Upload app.py with huggingface_hub

app.py

@@ -2,14 +2,21 @@ import gradio as gr
 from transformers import pipeline
 import torch
 
-# Load the NEW menu-aware model
+# Load the NEW menu-aware model (v2)
+# This model doesn't have pipeline.py - it's just a fine-tuned DistilBERT
 model_id = "samurai9776/thought-classifier-menu-aware"
+
+print(f"Loading model: {model_id}")
+
+# Load without trust_remote_code (new model doesn't need it)
 classifier = pipeline(
     "text-classification", 
     model=model_id,
     device=0 if torch.cuda.is_available() else -1
 )
 
+print("✅ Model loaded successfully!")
+
 def classify_thought(ai_utterance, cx_utterance):
     """Classify if the conversation is complete or incomplete"""
     
@@ -30,40 +37,37 @@ def classify_thought(ai_utterance, cx_utterance):
     # Determine prediction
     prediction = "Complete ✓" if complete_score > incomplete_score else "Incomplete ⚠️"
     
-    # Determine method used
+    # Determine confidence
     confidence = max(complete_score, incomplete_score)
-    if confidence > 0.95:
-        method = "High confidence prediction"
-    else:
-        method = "Moderate confidence prediction"
+    confidence_text = f"{confidence:.1%} confidence"
     
-    return prediction, complete_score, incomplete_score, method
+    return prediction, complete_score, incomplete_score, confidence_text
 
 # Create Gradio interface
 with gr.Blocks(
     title="Thought Completion Classifier - Menu Aware v2",
-    theme=gr.themes.Soft(),
-    css="""
-    .gradio-container {
-        font-family: 'IBM Plex Sans', sans-serif;
-    }
-    """
+    theme=gr.themes.Soft()
 ) as demo:
     gr.Markdown("""
     # 🤖 Thought Completion Classifier - Menu Aware v2
     
-    This enhanced model determines if a conversation represents a **complete** or **incomplete** thought.
-    Now with improved context understanding and menu-aware predictions!
+    Enhanced model with better context understanding and menu awareness.
     
-    ### 🎯 What's New:
+    ### ✨ What's New in v2:
+    - ✅ Trained on 15,000+ examples (vs 900 in v1)
     - ✅ Fixed 3,000+ mislabeled training examples
-    - ✅ Handles ambiguous terms (e.g., "chicken" could mean 75 different items!)
-    - ✅ Context-aware: "What sandwich?" + "Chicken" = Complete
-    - ✅ High precision for incomplete thought detection
+    - ✅ Handles ambiguous terms like "chicken" (75 menu items!)
+    - ✅ Better context understanding
+    - ✅ ~92% accuracy
+    
+    ### 🎯 How it works:
+    The model understands context:
+    - "What sandwich?" + "Chicken" = **Complete** (context is clear)
+    - "Anything else?" + "Chicken" = **Incomplete** (ambiguous)
     """)
     
     with gr.Row():
-        with gr.Column(scale=1):
+        with gr.Column():
             ai_input = gr.Textbox(
                 label="AI Utterance",
                 placeholder="e.g., What sandwich would you like?",
@@ -75,14 +79,13 @@ with gr.Blocks(
                 lines=2
             )
             
-            with gr.Row():
-                classify_btn = gr.Button("🔍 Classify", variant="primary", scale=2)
-                clear_btn = gr.Button("🗑️ Clear", variant="secondary", scale=1)
+            classify_btn = gr.Button("🔍 Classify", variant="primary", size="lg")
             
-        with gr.Column(scale=1):
+        with gr.Column():
             prediction = gr.Textbox(
                 label="Prediction", 
-                interactive=False
+                interactive=False,
+                elem_classes=["prediction-output"]
             )
             with gr.Row():
                 complete_score = gr.Number(
@@ -95,65 +98,46 @@ with gr.Blocks(
                     precision=3, 
                     interactive=False
                 )
-            method = gr.Textbox(
-                label="Confidence Level", 
+            confidence = gr.Textbox(
+                label="Confidence", 
                 interactive=False
             )
     
-    # Examples
-    gr.Markdown("### 📝 Try These Examples:")
+    # Examples showing context awareness
+    gr.Markdown("### 📝 Try these examples to see context awareness:")
     gr.Examples(
         examples=[
-            ["What can I get for you?", "Chicken"],
-            ["What sandwich would you like?", "Chicken"],
-            ["Chicken or beef?", "Chicken"],
-            ["Chicken or beef?", "Large"],
-            ["What size?", "Large"],
-            ["Anything else?", "Large"],
-            ["Ready to order?", "Spicy Chicken Sandwich Combo"],
-            ["Anything else?", "That's all"],
+            ["What can I get for you?", "Chicken", "Should be INCOMPLETE (ambiguous)"],
+            ["What sandwich would you like?", "Chicken", "Should be COMPLETE (context clear)"],
+            ["Chicken or beef?", "Chicken", "Should be COMPLETE (direct answer)"],
+            ["Chicken or beef?", "Large", "Should be INCOMPLETE (doesn't answer question)"],
+            ["What size?", "Large", "Should be COMPLETE (answers size question)"],
+            ["Anything else?", "Large", "Should be INCOMPLETE (size without item)"],
+            ["Ready to order?", "Spicy Chicken Sandwich Combo", "Should be COMPLETE (full item)"],
+            ["Anything else?", "That's all", "Should be COMPLETE (clear ending)"],
         ],
         inputs=[ai_input, cx_input],
-        outputs=[prediction, complete_score, incomplete_score, method],
+        outputs=[prediction, complete_score, incomplete_score, confidence],
         fn=classify_thought,
         cache_examples=True,
     )
     
-    # Event handlers
     classify_btn.click(
         fn=classify_thought,
         inputs=[ai_input, cx_input],
-        outputs=[prediction, complete_score, incomplete_score, method]
+        outputs=[prediction, complete_score, incomplete_score, confidence]
     )
     
-    clear_btn.click(
-        fn=lambda: ("", "", "", 0, 0, ""),
-        inputs=[],
-        outputs=[ai_input, cx_input, prediction, complete_score, incomplete_score, method]
-    )
+    gr.Markdown("""
+    ---
+    ### 📊 Model Comparison:
+    - **v1**: Rule-based + neural (samurai9776/thought-classifier)
+    - **v2**: Pure neural with better training (samurai9776/thought-classifier-menu-aware)
     
-    # Model info
-    with gr.Accordion("📊 Model Information", open=False):
-        gr.Markdown("""
-        ### Menu-Aware Model v2
-        - **Training samples:** 15,000+ (with 3,000+ corrections)
-        - **Accuracy:** ~92%
-        - **Base model:** DistilBERT
-        - **Special focus:** High precision for incomplete thoughts
-        
-        ### How it works:
-        1. **Context matters**: "What sandwich?" + "Chicken" = Complete
-        2. **Ambiguity detection**: "Anything else?" + "Chicken" = Incomplete
-        3. **Menu awareness**: Trained on 363 menu items with variations
-        
-        ### API Usage:
-        ```python
-        from transformers import pipeline
-        classifier = pipeline("text-classification", 
-                            model="samurai9776/thought-classifier-menu-aware")
-        result = classifier("What sandwich? [SEP] Chicken")
-        ```
-        """)
+    ### 🔗 Links:
+    - [Model v2 (Current)](https://huggingface.co/samurai9776/thought-classifier-menu-aware)
+    - [Model v1 (Previous)](https://huggingface.co/samurai9776/thought-classifier)
+    """)
 
 if __name__ == "__main__":
     demo.launch()