Update app.py

app.py

@@ -23,6 +23,8 @@ MAX_GPU_MEMORY = "60GiB"
 @GPU(memory=60)
 def load_model():
     """Load the Gemma 3 1B model without quantization for full precision."""
+    from transformers import pipeline
+    print(f"Attempting to load model: {MODEL_ID} without quantization")
     try:
         pipe = pipeline(
             "text-generation",
@@ -40,11 +42,10 @@ def load_model():
 @GPU(memory=60)
 def validate_request(name, context):
     """LLM-based request validation using isolated GPU function"""
-    try:
-        pipe = load_model()
-        
-        validation_prompt = [
-            {"role": "system", "content": """You are an ethical AI content moderator. Evaluate if this request is appropriate.
+    from transformers import pipeline  # Ensure pipeline is available in GPU process
+    
+    validation_prompt = [
+        {"role": "system", "content": """You are an ethical AI content moderator. Evaluate if this request is appropriate.
 CRITERIA:
 1. NO minors (under 18) or underage references
 2. NO vulnerable populations
@@ -52,9 +53,11 @@ CRITERIA:
 4. NO illegal/harmful scenarios
 5. NO inappropriate relationships
 Respond ONLY with "TRUE" if acceptable, "FALSE" if not."""},
-            {"role": "user", "content": f"Character Name: {name}\nContext: {context}"}
-        ]
-        
+        {"role": "user", "content": f"Character Name: {name}\nContext: {context}"}
+    ]
+    
+    try:
+        pipe = load_model()
         tokenizer = pipe.tokenizer
         text = tokenizer.apply_chat_template(
             validation_prompt,
@@ -117,6 +120,53 @@ def search_person(name, context=""):
     print(f"Found {len(results)} potential search results.")
     return results
 
+def create_synthetic_profile(name, context):
+    """Create a synthetic profile when search returns no results."""
+    profile = {
+        "title": f"Synthetic Profile for {name}",
+        "href": "",
+        "body": f"{name} is a person described with the context: '{context}'. "
+    }
+    if "grade" in context.lower():
+        grade_match = re.search(r'(\d+)(?:st|nd|rd|th)?\s+grade', context.lower())
+        if grade_match:
+            try:
+                grade = int(grade_match.group(1))
+                age = 5 + grade
+                profile["body"] += f"Based on being in {grade}th grade, {name} is likely around {age} years old. "
+                profile["body"] += f"Typical interests for this age might include friends, hobbies, school subjects, and developing independence. "
+            except ValueError:
+                profile["body"] += f"The grade mentioned ('{grade_match.group(1)}') could not be parsed to estimate age. "
+    profile["body"] += "Since no public information was found, this profile is based solely on the provided context."
+    return [profile]
+
+def extract_text_from_search_results(search_results):
+    """Extract relevant text from search results."""
+    if isinstance(search_results, str):
+        return f"Could not extract text due to search error: {search_results}"
+    
+    combined_text = ""
+    seen_bodies = set()
+    count = 0
+    max_results_to_process = 5
+    
+    for result in search_results:
+        if count >= max_results_to_process:
+            break
+        if isinstance(result, dict) and 'body' in result and result['body']:
+            body = result['body'].strip()
+            if body not in seen_bodies:
+                combined_text += body + "\n"
+                seen_bodies.add(body)
+                count += 1
+    
+    if not combined_text:
+        return "No relevant text found in search results."
+    
+    combined_text = re.sub(r'\s+', ' ', combined_text).strip()
+    max_length = 2000
+    return combined_text[:max_length] + "..." if len(combined_text) > max_length else combined_text
+
 # --- Text Processing Functions ---
 def parse_llm_output(full_output, input_prompt_list):
     """Attempts to parse only the newly generated text from the LLM output."""
@@ -165,6 +215,8 @@ def parse_llm_output(full_output, input_prompt_list):
 @GPU(memory=60)
 def generate_enhanced_persona(name, bio_text, context=""):
     """Use the LLM to enhance the persona profile."""
+    from transformers import pipeline
+    
     pipe = load_model()
     print(f"Generating enhanced persona for {name}...")
     
@@ -203,6 +255,8 @@ Create the profile based *only* on the text above."""}
 @GPU(memory=60)
 def generate_system_prompt_with_llm(name, enhanced_profile, context=""):
     """Generate an optimized system prompt for the persona."""
+    from transformers import pipeline
+    
     pipe = load_model()
     print(f"Generating system prompt for {name}...")
     
@@ -213,11 +267,7 @@ Additional context for the simulation: {context}
 Maintain this persona consistently. Respond naturally based on the profile. Do not mention that you are an AI or a simulation. If asked about details not in the profile, you can be evasive or state you don't know/remember, consistent with the persona."""
     
     prompt = [
-        {"role": "system", "content": """You are an expert AI prompt engineer specializing in character simulation. Create a concise system prompt that instructs the LLM to embody the character based on the profile. The prompt must:
-1. Define core personality and speaking style
-2. Specify how to handle unknown topics
-3. Prohibit breaking character or mentioning AI nature
-Output ONLY the system prompt itself."""
+        {"role": "system", "content": """You are an expert AI prompt engineer specializing in character simulation. Create a concise system prompt that instructs the LLM to embody the character based on the profile. The prompt must: 1. Define core personality and speaking style. 2. Specify how to handle unknown topics. 3. Prohibit breaking character or mentioning AI nature. Output ONLY the system prompt itself."""
         },
         {"role": "user", "content": f"""Create a system prompt for an AI to simulate the character '{name}'. Context for simulation: {context} Character Profile:
 {enhanced_profile}
@@ -251,6 +301,8 @@ Generate the system prompt based *only* on the profile and context provided."""}
 @GPU(memory=60)
 def generate_response(messages):
     """Generate a response using the LLM."""
+    from transformers import pipeline
+    
     pipe = load_model()
     print("Generating response...")
     if not messages:
@@ -279,7 +331,7 @@ def generate_response(messages):
     except Exception as e:
         error_msg = f"Error during response generation: {str(e)}"
         print(error_msg)
-        return f"Sorry, I encountered an error: {str(e)}"
+        return f"Sorry, I encountered an error trying to respond."
 
 # --- Persona Chat Class with Safety ---
 class PersonaChat:
@@ -466,7 +518,6 @@ def create_interface():
                     chatbot = gr.Chatbot(
                         height=450, 
                         show_label=False,
-                        bubble_full_width=False,
                         type="messages",
                         avatar_images=("https://api.dicebear.com/6.x/bottts/svg?seed=user ", 
                                       "https://api.dicebear.com/6.x/bottts/svg?seed=bot ")
@@ -506,7 +557,7 @@ def create_interface():
                     time.sleep(0.1)
             except Exception as e:
                 error_msg = f"Failed to set persona (interface error): {str(e)}"
-                print(error_msg)
+                print(f"set_persona_flow: Exception: {error_msg}")
                 yield error_msg, "", "", f"### Error creating {name}", []
         
         def send_message_flow(message, history):