Update app.py

app.py

@@ -2,16 +2,9 @@ import gradio as gr
 from huggingface_hub import InferenceClient
 
 # --- Configuration ---
-# OPTION 1: Qwen 2.5 Coder (Recommended: Faster, Smarter)
 MODEL_QWEN = "Qwen/Qwen2.5-Coder-7B-Instruct"
-
-# OPTION 2: CodeGeeX4 (GLM-4 Architecture)
-# Note: This might timeout more often on the free tier because it's larger (9B)
 MODEL_GLM = "THUDM/codegeex4-all-9b"
 
-# We default to Qwen because it is more reliable on the Free API
-CURRENT_MODEL = MODEL_QWEN
-
 def generate_abap(message, history, model_choice):
     # Select the model based on user dropdown
     if model_choice == "GLM-4 (CodeGeeX4)":
@@ -20,23 +13,21 @@ def generate_abap(message, history, model_choice):
         model_id = MODEL_QWEN
         
     client = InferenceClient(model=model_id)
-
+    
     # System Prompt specialized for ABAP
     system_prompt = "You are an expert SAP ABAP Developer. Write modern, efficient ABAP 7.4+ code. Always use inline declarations."
     
     # Construct Prompt
-    # We use a generic chat format that works for both models
     prompt = f"System: {system_prompt}\n"
     for user, bot in history:
         prompt += f"User: {user}\nAssistant: {bot}\n"
     prompt += f"User: {message}\nAssistant:"
 
     try:
-        # Stream response from HF GPU
         stream = client.text_generation(
             prompt,
             max_new_tokens=1024,
-            temperature=0.1, # Precise code
+            temperature=0.1,
             top_p=0.9,
             stream=True,
             details=True,
@@ -46,35 +37,33 @@ def generate_abap(message, history, model_choice):
         partial_message = ""
         for response in stream:
             token = response.token.text
-            # Filter out stop tokens
             if token not in ["User:", "System:"]:
                 partial_message += token
                 yield partial_message
                 
     except Exception as e:
-        error_msg = f"Error: The Free API is currently overloaded for {model_id}. \n\nTechnical details: {str(e)}"
-        yield error_msg
+        yield f"Error: The Free API is overloaded. Details: {str(e)}"
 
 # --- The UI ---
 with gr.Blocks( ) as demo:
     gr.Markdown("# 🚀 ABAP Coder (Serverless GPU)")
     gr.Markdown("Generate ABAP code using top open-source models running on Hugging Face's Free API.")
     
-    # Dropdown to choose model
     model_selector = gr.Dropdown(
         choices=["Qwen 2.5 Coder (Recommended)", "GLM-4 (CodeGeeX4)"],
         value="Qwen 2.5 Coder (Recommended)",
         label="Select AI Model"
     )
     
-    # Chat Interface
     chat = gr.ChatInterface(
         fn=generate_abap,
         additional_inputs=[model_selector],
+        # --- THE FIX IS HERE ---
+        # Each example is now a list: [Question, Dropdown Value]
         examples=[
-            "Write a report to select data from MARA using inline declarations.",
-            "Create a CDS View for Sales Orders (VBAK/VBAP).",
-            "Explain how to use READ TABLE with ASSIGNING FIELD-SYMBOL."
+            ["Write a report to select data from MARA using inline declarations.", "Qwen 2.5 Coder (Recommended)"],
+            ["Create a CDS View for Sales Orders (VBAK/VBAP).", "Qwen 2.5 Coder (Recommended)"],
+            ["Explain how to use READ TABLE with ASSIGNING FIELD-SYMBOL.", "Qwen 2.5 Coder (Recommended)"]
         ]
     )