Update app.py

app.py

@@ -1,164 +1,79 @@
 import os
 import openai
 import gradio as gr
-from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
-import torch
 
 # Load OpenAI API key
 openai.api_key = os.getenv("OPENAI_API_KEY")
 
-# Try to define the fallback model ahead of time so it's ready
-llama_model = None
-llama_tokenizer = None
+llama_response = None  # Global for LLaMA fallback model
 
 def call_openai(prompt):
-    """Call OpenAI API to analyze the prompt rather than just answering it."""
+    """Call OpenAI GPT-3.5 to analyze the prompt."""
     try:
-        # Detailed system prompt for prompt coaching rather than answering
         system_prompt = """You are a helpful Prompt Interpretation Coach.
-        
-        Your goal is to:
-        1. Explain how an AI would interpret their prompt (not to answer it directly)
-        2. Highlight any ambiguities or potential misunderstandings in their prompt
-        3. Suggest improvements to make their prompt clearer and more effective
-        4. Offer 1-2 alternative phrasings that might work better
-        
-        EXTREMELY IMPORTANT RULES:
-        - NEVER actually execute or fulfill the student's request
-        - For example, if they ask "Write a poem about AI," DO NOT write a poem - instead analyze how AI would interpret that prompt
-        - Begin your response with "PROMPT ANALYSIS:" to reinforce you are analyzing, not answering
-        - Keep your language and themes appropriate for high school students and young kids
-        - Avoid any harmful, inappropriate, or mature content
-        - Focus on educational value and critical thinking skills
-        - Be encouraging and constructive in your feedback
-        - Keep responses concise (max 4-5 short paragraphs)
-        - If the prompt asks for potentially harmful content, gently redirect without repeating harmful elements
-        """
-        
-        # Updated to use current OpenAI client library
+
+Your goal is to:
+1. Explain how an AI would interpret their prompt (not to answer it directly)
+2. Highlight any ambiguities or potential misunderstandings in their prompt
+3. Suggest improvements to make their prompt clearer and more effective
+4. Offer 1-2 alternative phrasings that might work better
+
+EXTREMELY IMPORTANT RULES:
+- NEVER actually execute or fulfill the student's request
+- Begin your response with "PROMPT ANALYSIS:"
+- Keep your tone appropriate for young learners
+- Be clear, constructive, and concise (4-5 short paragraphs max)
+"""
         client = openai.OpenAI(api_key=openai.api_key)
         response = client.chat.completions.create(
             model="gpt-3.5-turbo",
             messages=[
                 {"role": "system", "content": system_prompt},
-                {"role": "user", "content": f"Provide ONLY an analysis of this prompt: \"{prompt}\". DO NOT fulfill the request, only analyze it."}
-            ],
-            temperature=0.7,
-            max_tokens=350
-        )
-        return response.choices[0].message.content.strip()
-    except AttributeError:
-        # Fallback for older version of OpenAI client
-        response = openai.ChatCompletion.create(
-            model="gpt-3.5-turbo",
-            messages=[
-                {"role": "system", "content": system_prompt},
-                {"role": "user", "content": f"Analyze this prompt: \"{prompt}\""}
+                {"role": "user", "content": f"Analyze this prompt ONLY: \"{prompt}\""}
             ],
             temperature=0.7,
             max_tokens=350
         )
         return response.choices[0].message.content.strip()
 
+    except Exception as e:
+        raise e  # Bubble up to interpret_prompt
+
 def call_llama(prompt):
-    """Call LLaMA model to analyze the prompt rather than just answering it."""
-    global llama_model, llama_tokenizer
-    if llama_model is None or llama_tokenizer is None:
+    """Fallback to LLaMA hosted on Hugging Face via novita."""
+    global llama_response
+    if llama_response:
         try:
-            llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct")
-            llama_model = AutoModelForCausalLM.from_pretrained(
-                "meta-llama/Meta-Llama-3-8B-Instruct", 
-                torch_dtype=torch.float16, 
-                device_map="auto"
-            )
+            return llama_response(prompt)
         except Exception as e:
-            return f"Failed to load LLaMA model: {str(e)}"
-    
-    try:
-        # Create a more detailed prompt for LLaMA to analyze rather than answer
-        full_prompt = f"""<|system|>
-You are a Prompt Interpretation Coach. Your ONLY job is to analyze how an AI would interpret the following prompt.
-
-EXTREMELY IMPORTANT RULES:
-1. NEVER execute or fulfill the actual request in the prompt
-2. ALWAYS start your response with "PROMPT ANALYSIS:" to emphasize you are analyzing
-3. If the prompt says "Write a poem about AI" - DO NOT write a poem! Instead explain how an AI would interpret that request
-4. Explain how an AI system would understand this prompt
-5. Point out any unclear parts or potential misunderstandings
-6. Suggest specific improvements to make the prompt more effective
-7. Always keep content appropriate for young learners
-8. Be encouraging and educational
-
-Keep your analysis concise (4-5 short paragraphs maximum).
-<|user|>
-ONLY analyze this prompt (DO NOT fulfill it): "{prompt}"
-<|assistant|>
-PROMPT ANALYSIS:
-"""
-        
-        inputs = llama_tokenizer(full_prompt, return_tensors="pt")
-        # Move to the device where the model is
-        if torch.cuda.is_available():
-            inputs = inputs.to("cuda")
-        else:
-            inputs = inputs.to(llama_model.device)
-            
-        outputs = llama_model.generate(
-            **inputs, 
-            max_new_tokens=350,
-            temperature=0.7,
-            do_sample=True
-        )
-        
-        generated_text = llama_tokenizer.decode(outputs[0], skip_special_tokens=True)
-        
-        # Extract just the assistant's response if possible
-        if "<|assistant|>" in generated_text:
-            return generated_text.split("<|assistant|>")[1].strip()
-        return generated_text
-    except Exception as e:
-        return f"Error generating response with LLaMA: {str(e)}"
+            return f"⚠️ Error using LLaMA fallback: {str(e)}"
+    return "⚠️ LLaMA fallback not available. Please sign in with Hugging Face in the sidebar."
 
 def interpret_prompt(prompt):
-    """Main function to interpret and analyze the prompt."""
-    # Check if prompt is empty or too short
+    """Main logic to analyze prompt quality and clarity."""
     if not prompt or len(prompt.strip()) < 3:
-        return "Please enter a prompt to analyze. Try something like 'Write a poem about space' or 'Explain quantum physics simply'."
-        
-    # Content safety check - very basic implementation
-    # A more robust implementation would use a dedicated content filter
-    unsafe_terms = [
-        "suicide", "self-harm", "kill", "porn", "naked", "nude", "sexual", 
-        "weapon", "bomb", "terrorist", "hack", "steal", "drug", "cocaine", 
-        "heroin", "illegal", "torrent", "pirate"
-    ]
-    
-    # Check for unsafe content
-    lower_prompt = prompt.lower()
-    for term in unsafe_terms:
-        if term in lower_prompt:
-            return "⚠️ I notice your prompt might be asking for content that isn't appropriate for this educational tool. This coach is designed to help with academic and creative prompts. Could you try a different prompt related to school projects, creative writing, or learning concepts?"
-    
-    # Add a prefix to the output to further emphasize we're analyzing, not answering
+        return "Please enter a prompt to analyze. Try something like 'Write a poem about space'."
+
+    # Simple safety check
+    unsafe_terms = ["suicide", "self-harm", "kill", "porn", "naked", "nude", "sexual", "weapon", "bomb", "terrorist", "hack", "steal", "drug", "cocaine", "heroin", "illegal", "torrent", "pirate"]
+    if any(term in prompt.lower() for term in unsafe_terms):
+        return "⚠️ That topic isn't appropriate for this tool. Try a school project, creative writing, or learning-related prompt."
+
     prefix = "AI Response:\n\n"
-    
+
     try:
         response = call_openai(prompt)
-        # If response starts with a common answer pattern instead of an analysis,
-        # add an override message
-        if any(response.lower().startswith(x) for x in ["here's a poem", "here is a poem", "once upon", "in a"]):
-            return prefix + "⚠️ Note: It appears the AI tried to fulfill your request rather than analyze it. This is exactly why prompt engineering is important! Let me provide the correct analysis instead:\n\n" + call_openai("I need you to ONLY analyze this prompt and NOT fulfill it: " + prompt)
+        if any(response.lower().startswith(x) for x in ["here's a poem", "once upon", "in a world", "roses are red"]):
+            return prefix + "⚠️ The AI started fulfilling your request. Let's try again with clearer instructions:\n\n" + call_openai("ONLY analyze this prompt: " + prompt)
         return prefix + response
     except Exception as e:
-        error_str = str(e).lower()
-        if "rate limit" in error_str:
-            return prefix + "[⚠️ OpenAI Rate Limit Hit — switching to LLaMA 3...]\n\n" + call_llama(prompt)
-        elif "auth" in error_str or "api key" in error_str:
+        if "rate limit" in str(e).lower():
+            return prefix + "[⚠️ OpenAI Rate Limit Hit — switching to LLaMA fallback...]\n\n" + call_llama(prompt)
+        elif "auth" in str(e).lower() or "api key" in str(e).lower():
             return "❌ Invalid or missing OpenAI API key."
-        else:
-            return prefix + f"Unexpected error: {str(e)}\n\nFalling back to LLaMA model...\n\n" + call_llama(prompt)
+        return prefix + f"⚠️ Unexpected error: {str(e)}\n\nSwitching to LLaMA fallback...\n\n" + call_llama(prompt)
 
-# Create a custom theme with better visuals
+# Theme
 custom_theme = gr.themes.Soft(
     primary_hue="indigo",
     secondary_hue="blue",
@@ -166,68 +81,59 @@ custom_theme = gr.themes.Soft(
     radius_size=gr.themes.sizes.radius_sm,
 )
 
-# Updated interface with improved visuals and clarity
-iface = gr.Interface(
-    fn=interpret_prompt,
-    inputs=gr.Textbox(
-        lines=3, 
-        placeholder="Enter a prompt you'd like to analyze (e.g., 'Draw a star with turtle graphics')...",
-        elem_id="prompt-input"
-    ),
-    outputs=gr.Textbox(
-        label="Prompt Analysis & Coaching Tips",
-        elem_id="analysis-output"
-    ),
-    title="🧠 Prompt Interpretation Coach",
-    description="""## What is Prompt Interpretation Coach?
-This educational tool helps leaners learn **prompt engineering** by analyzing how AI systems interpret instructions.
-
-### ⚙ How it works:
-* **Enter any prompt** you might give to an AI system
-* The coach **analyzes your prompt** instead of answering it
-* Learn what makes an effective prompt through real examples
-* Develop critical thinking about AI communication
-
-### 📚 What you'll learn:
-* How AI would understand your instructions
-* Potential ambiguities or misunderstandings
-* Specific improvements for clearer prompts
-* Alternative phrasings that might work better
-
-*Try the examples below or enter your own prompt to start learning!*
-    """,
-    examples=[
-        "Draw a star with turtle graphics",
-        "Write a poem about AI",
-        "Explain photosynthesis",
-        "Plan my next trip",
-        "Explain algebra like I'm 10"
-    ],
-    theme=custom_theme,
-    elem_id="prompt-coach-interface"
-)
-
-# Add footer with attribution
-footer_html = """## ⌨️ Why Prompt Engineering Matters
+# Gradio UI
+with gr.Blocks(theme=custom_theme, fill_height=True) as demo:
+    with gr.Sidebar():
+        gr.Markdown("## 🧠 Prompt Coach Settings")
+        gr.Markdown("OpenAI is used by default.\nIf rate-limited, fallback to LLaMA 3 via Hugging Face.")
+        login_button = gr.LoginButton("🔐 Sign in to use LLaMA fallback")
+
+    # Load LLaMA from Hugging Face using Novita provider
+    llama_response = gr.load(
+        "models/meta-llama/Meta-Llama-3-8B-Instruct",
+        provider="novita",
+        accept_token=login_button
+    )
+
+    gr.Interface(
+        fn=interpret_prompt,
+        inputs=gr.Textbox(
+            lines=3,
+            placeholder="e.g. 'Draw a star with turtle graphics'",
+            elem_id="prompt-input"
+        ),
+        outputs=gr.Textbox(
+            label="Prompt Analysis & Coaching Tips",
+            elem_id="analysis-output"
+        ),
+        title="🧠 Prompt Interpretation Coach",
+        description="""This tool helps students and creators understand how AI interprets their prompts — without actually answering them.
+
+### What you’ll learn:
+- How AI interprets your instructions
+- Ambiguities in your phrasing
+- Better alternatives for clearer prompts
+- Prompt engineering strategies for education
+
+Try typing a school or creative prompt to begin!""",
+        examples=[
+            "Draw a star with turtle graphics",
+            "Write a poem about AI",
+            "Explain photosynthesis",
+            "Plan my next trip",
+            "Explain algebra like I'm 10"
+        ],
+        elem_id="prompt-coach-interface"
+    ).render()
+
+    gr.Markdown("""## ⌨️ Why Prompt Engineering Matters
 Learning to craft clear, specific prompts helps you:
+- 📚 In education: Get more accurate responses
+- 💡 In creativity: Express ideas more clearly
+- 🧠 In learning: Improve critical thinking
 
-* **In education:** Get more accurate and helpful responses for homework and projects
-* **In research:** Direct AI to explore specific aspects of a topic
-* **In creativity:** Guide AI to help with your creative vision more effectively
-* **In coding:** Get more precise code suggestions and explanations
-
-**Made with love ❤️ for learners exploring the world of AI + code.**  
-Created by Shingai Manjengwa, @tjido
-"""
-
-# Create the Gradio app with the footer
-demo = gr.Blocks()
-with demo:
-    # gr.Image("thumbnail.png", show_label=False, height=100, width=100)
-    iface.render()
-    gr.Markdown(footer_html)
+**Created with ❤️ by Shingai Manjengwa | @tjido**
+""")
 
 if __name__ == "__main__":
     demo.launch()
-
-