Upload app.py

app.py

@@ -359,80 +359,137 @@ def analyze_solution(question: str, solution: str):
      
 
 
-
-def classify_solution(question: str, solution: str):
-    """
-    Classify the math solution
-    Returns: (classification_label, confidence_score, explanation)
-    """
+def classify_solution_stream(question: str, solution: str, progress=gr.Progress()):
+    # always yield EXACTLY THREE outputs (classification, explanation, status)
     if not question.strip() or not solution.strip():
-        return "Please fill in both fields", 0.0, ""
-    
+        yield "Please fill in both fields", "", "⚠️ Provide a question and a solution."
+        return
+
+    # lazy init if needed
     if not models_ready():
-        return "Models not loaded", 0.0, ""
-    
+        yield "⏳ Loading models…", "", "Booting models…"
+        msg = load_model()
+        progress(0.1, desc="Loading models")
+        if not models_ready():
+            yield "Models not loaded", "", f"❌ {msg}"
+            return
+
     try:
-        res = analyze_solution(question, solution)
-        
-        return res["classification"], res["explanation"]
-    except Exception:
+        # Stage 1 — conceptual
+        progress(0.2, desc="Stage 1: Conceptual check")
+        yield "⏳ Working…", "Starting conceptual check…", "🔎 **Stage 1:** running classifier…"
+        conceptual = run_conceptual_check(question, solution, classifier_model, classifier_tokenizer)
+        conf = conceptual['probabilities'][conceptual['prediction']]
+        yield "⏳ Working…", f"Stage 1: model predicts **{conceptual['prediction']}** (confidence {conf:.2%}). Now checking calculations…", \
+              f"✅ **Stage 1 done** — prediction: **{conceptual['prediction']}** (p={conf:.2%})."
+
+        # Stage 2 — computational
+        progress(0.6, desc="Stage 2: Computational check")
+        yield "⏳ Working…", "Running computational check…", "🧮 **Stage 2:** extracting & evaluating equations…"
+        computational = run_computational_check(solution, gemma_model, gemma_tokenizer)
+
+        # Final verdict
+        if computational["error"]:
+            classification = "computational_error"
+            explanation = (
+                f"A calculation error was found.\n"
+                f"On the line: \"{computational['line_text']}\"\n"
+                f"The correct calculation should be: {computational['correct_calc']}"
+            )
+            status = "🟥 **Final:** computational error detected."
+        else:
+            if conceptual['prediction'] == 'correct':
+                classification = 'correct'
+                explanation = "All calculations are correct and the overall logic appears to be sound."
+                status = "🟩 **Final:** correct."
+            else:
+                classification = 'conceptual_error'
+                explanation = "All calculations are correct, but there appears to be a conceptual error in the logic or setup of the solution."
+                status = "🟨 **Final:** conceptual issue."
+
+        progress(1.0, desc="Done")
+        yield classification, explanation, status
+
+    except Exception as e:
         logger.exception("inference failed")
+        yield "Runtime error", f"{type(e).__name__}: {e}", "❌ Exception during inference."
+
         
         
         
 
-
-
-# Create Gradio interface
+# Create Gradio interface (streaming UI)
 with gr.Blocks(title="Math Solution Classifier", theme=gr.themes.Soft()) as app:
     gr.Markdown("# 🧮 Math Solution Classifier")
-    gr.Markdown("Classify math solutions as correct, conceptually flawed, or computationally flawed.")
-    
+    gr.Markdown("Classify math solutions as **correct**, **conceptually flawed**, or **computationally flawed**. "
+                "Status updates stream below as the two-stage pipeline runs.")
+
     with gr.Row():
-        with gr.Column():
+        # ---------- Left: inputs ----------
+        with gr.Column(scale=1):
             question_input = gr.Textbox(
                 label="Math Question",
                 placeholder="e.g., Solve for x: 2x + 5 = 13",
-                lines=3
+                lines=3,
             )
-            
             solution_input = gr.Textbox(
-                label="Proposed Solution", 
+                label="Proposed Solution",
                 placeholder="e.g., 2x + 5 = 13\n2x = 13 - 5\n2x = 8\nx = 4",
-                lines=5
+                lines=8,
             )
-            
-            classify_btn = gr.Button("Classify Solution", variant="primary")
-        
-        with gr.Column():
+            with gr.Row():
+                classify_btn = gr.Button("Classify Solution", variant="primary")
+                clear_btn = gr.Button("Clear")
+
+        # ---------- Right: outputs ----------
+        with gr.Column(scale=1):
             classification_output = gr.Textbox(label="Classification", interactive=False)
-            confidence_output = gr.Textbox(label="Confidence", interactive=False)
-            explanation_output = gr.Textbox(label="Explanation", interactive=False, lines=3)
-    
-    # Examples
+            explanation_output = gr.Textbox(label="Explanation", interactive=False, lines=6)
+            status_output = gr.Markdown(value="*(idle)*")  # live stage updates
+
+    # ---------- Examples ----------
     gr.Examples(
         examples=[
             [
                 "Solve for x: 2x + 5 = 13",
-                "2x + 5 = 13\n2x = 13 - 5\n2x = 8\nx = 4"
+                "2x + 5 = 13\n2x = 13 - 5\n2x = 8\nx = 4",
             ],
             [
-                "John has three apples and Mary has seven, how many apples do they have together?", 
-                "They have 7 + 3 = 11 apples."  # This should be computationally flawed
+                "John has three apples and Mary has seven, how many apples do they have together?",
+                "They have 7 + 3 = 11 apples.",
             ],
             [
                 "What is 15% of 200?",
-                "15% = 15/100 = 0.15\n0.15 × 200 = 30"
-            ]
+                "15% = 15/100 = 0.15\n0.15 × 200 = 30",
+            ],
         ],
-        inputs=[question_input, solution_input]
+        inputs=[question_input, solution_input],
     )
-    
+
+    # ---------- Wiring ----------
     classify_btn.click(
-        fn=classify_solution,
+        fn=classify_solution_stream,  # <- your generator function
         inputs=[question_input, solution_input],
-        outputs=[classification_output, explanation_output]
+        outputs=[classification_output, explanation_output, status_output],
+        show_progress="full",
     )
 
+    # Clear everything
+    clear_btn.click(
+        lambda: ("", "", "", "", "*(idle)*"),
+        inputs=None,
+        outputs=[
+            question_input,
+            solution_input,
+            classification_output,
+            explanation_output,
+            status_output,
+        ],
+        queue=False,
+    )
+
+if __name__ == "__main__":
+    app.launch()
+
 if __name__ == "__main__":
     app.launch()