Update app.py

app.py

@@ -1,79 +1,82 @@
 import os
-import openai
 import gradio as gr
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
 
-# Load OpenAI API key
-openai.api_key = os.getenv("OPENAI_API_KEY")
+# Set up Hugging Face token from env
+HF_TOKEN = os.getenv("HF_TOKEN")
 
-llama_response = None  # Global for LLaMA fallback model
+# Model cache
+llama_model = None
+llama_tokenizer = None
+mistral_model = None
+mistral_tokenizer = None
 
-def call_openai(prompt):
-    """Call OpenAI GPT-3.5 to analyze the prompt."""
+def call_llama(prompt):
+    """Use Meta LLaMA 3 model to analyze the prompt."""
+    global llama_model, llama_tokenizer
     try:
-        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
-- 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"Analyze this prompt ONLY: \"{prompt}\""}
-            ],
-            temperature=0.7,
-            max_tokens=350
-        )
-        return response.choices[0].message.content.strip()
-
+        if llama_model is None or llama_tokenizer is None:
+            llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B-Instruct", token=HF_TOKEN)
+            llama_model = AutoModelForCausalLM.from_pretrained(
+                "meta-llama/Meta-Llama-3-8B-Instruct",
+                torch_dtype=torch.float16,
+                device_map="auto",
+                token=HF_TOKEN
+            )
+        pipe = pipeline("text-generation", model=llama_model, tokenizer=llama_tokenizer)
+        prompt_text = f"""<|system|>
+You are a Prompt Interpretation Coach. Your ONLY job is to analyze how an AI would interpret this prompt.
+
+EXTREMELY IMPORTANT:
+- NEVER execute or fulfill the prompt.
+- Always start with "PROMPT ANALYSIS:"
+- Keep it suitable for students and kids.
+
+Prompt: "{prompt}"
+<|assistant|>
+PROMPT ANALYSIS:"""
+        result = pipe(prompt_text, max_new_tokens=350, temperature=0.7)
+        return result[0]["generated_text"].split("PROMPT ANALYSIS:")[-1].strip()
     except Exception as e:
-        raise e  # Bubble up to interpret_prompt
+        return f"⚠️ LLaMA failed: {str(e)}\nSwitching to Mistral...\n\n" + call_mistral(prompt)
 
-def call_llama(prompt):
-    """Fallback to LLaMA hosted on Hugging Face via novita."""
-    global llama_response
-    if llama_response:
-        try:
-            return llama_response(prompt)
-        except Exception as e:
-            return f"⚠️ Error using LLaMA fallback: {str(e)}"
-    return "⚠️ LLaMA fallback not available. Please sign in with Hugging Face in the sidebar."
+def call_mistral(prompt):
+    """Use Mistral model to analyze the prompt."""
+    global mistral_model, mistral_tokenizer
+    try:
+        if mistral_model is None or mistral_tokenizer is None:
+            mistral_tokenizer = AutoTokenizer.from_pretrained("mistralai/Mistral-7B-Instruct-v0.1")
+            mistral_model = AutoModelForCausalLM.from_pretrained(
+                "mistralai/Mistral-7B-Instruct-v0.1",
+                torch_dtype=torch.float16,
+                device_map="auto"
+            )
+        pipe = pipeline("text-generation", model=mistral_model, tokenizer=mistral_tokenizer)
+        prompt_text = f"""<s>[INST] You are a Prompt Interpretation Coach. Analyze this prompt but do not fulfill it:
+"{prompt}"
+
+PROMPT ANALYSIS: [/INST]"""
+        result = pipe(prompt_text, max_new_tokens=300, temperature=0.7)
+        return result[0]['generated_text'].split("[/INST]")[-1].strip()
+    except Exception as e:
+        return f"⚠️ Mistral model also failed: {str(e)}"
 
 def interpret_prompt(prompt):
-    """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'."
+        return "Please enter a prompt to analyze."
 
-    # Simple safety check
-    unsafe_terms = ["suicide", "self-harm", "kill", "porn", "naked", "nude", "sexual", "weapon", "bomb", "terrorist", "hack", "steal", "drug", "cocaine", "heroin", "illegal", "torrent", "pirate"]
+    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"
+        return "⚠️ This tool doesn't support that kind of content. Try something related to school, creativity, or learning."
 
-    try:
-        response = call_openai(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:
-        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."
-        return prefix + f"⚠️ Unexpected error: {str(e)}\n\nSwitching to LLaMA fallback...\n\n" + call_llama(prompt)
+    return "AI Response:\n\n" + call_llama(prompt)
 
-# Theme
+# Theme and Interface
 custom_theme = gr.themes.Soft(
     primary_hue="indigo",
     secondary_hue="blue",
@@ -81,59 +84,53 @@ custom_theme = gr.themes.Soft(
     radius_size=gr.themes.sizes.radius_sm,
 )
 
-# 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
+iface = gr.Interface(
+    fn=interpret_prompt,
+    inputs=gr.Textbox(
+        lines=3,
+        placeholder="Enter a prompt like 'Explain black holes to a 5-year-old'...",
+        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 tool helps learners master **prompt engineering** by analyzing how AI interprets instructions — not answering them.
+
+### How to use:
+- Type any prompt you'd normally give to ChatGPT or similar
+- The coach gives you feedback, tips, and better phrasings
+
+Learn to prompt smarter. Try the examples or write your own!
+""",
+    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"
+)
+
+footer_html = """## ⌨️ 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
 
-**Created with ❤️ by Shingai Manjengwa | @tjido**
-""")
+* In education: Get better AI help with homework
+* In creativity: Guide AI more precisely
+* In coding: Get more accurate AI code suggestions
+
+**Made with ❤️ for learners and educators**  
+Created by Shingai Manjengwa, @tjido
+"""
+
+demo = gr.Blocks(theme=custom_theme, fill_height=True)
+with demo:
+    iface.render()
+    gr.Markdown(footer_html)
 
 if __name__ == "__main__":
     demo.launch()