app/main.py

import os
import torch
from fastapi import FastAPI, Request
from fastapi.responses import JSONResponse
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
from starlette.middleware.cors import CORSMiddleware

# === Setup FastAPI ===
app = FastAPI(title="Apollo AI Backend - Qwen2-0.5B", version="3.1.0-FIXED")

# === CORS ===
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# === Configuration ===
API_KEY = os.getenv("API_KEY", "aigenapikey1234567890")
BASE_MODEL = "Qwen/Qwen2-0.5B-Instruct"
ADAPTER_PATH = "adapter"

# === Load Model ===
print("🔧 Loading tokenizer for Qwen2-0.5B...")
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

print("🧠 Loading Qwen2-0.5B base model...")
base_model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    trust_remote_code=True,
    torch_dtype=torch.float32,
    device_map="cpu"
)

print("🔗 Applying LoRA adapter to Qwen2-0.5B...")
model = PeftModel.from_pretrained(base_model, ADAPTER_PATH)
model.eval()

print("✅ Qwen2-0.5B model ready!")

def create_conversation_prompt(messages: list, is_force_mode: bool) -> str:
    """
    Create a conversation prompt with STRONG mode enforcement
    """
    if is_force_mode:
        system_prompt = """FORCE MODE - DIRECT ANSWERS ONLY:
You MUST give direct, complete, factual answers. Do NOT ask questions. Provide exact solutions, working code, and clear explanations.

EXAMPLE FORCE RESPONSE:
Q: What does len() do in Python?
A: len() returns the number of items in an object. Examples:
- len([1,2,3]) returns 3
- len("hello") returns 5  
- len({1,2,3}) returns 3

Always be direct and informative. Never ask "What do you think?" or similar questions."""
    else:
        system_prompt = """MENTOR MODE - GUIDED LEARNING ONLY:
You are a programming teacher. You MUST guide students to discover answers themselves. NEVER give direct answers or complete solutions. ALWAYS respond with guiding questions and hints.

EXAMPLE MENTOR RESPONSE:
Q: What does len() do in Python?
A: Great question! What do you think might happen if you run len([1,2,3]) in Python? Can you guess what number it would return? Try it and see! What pattern do you notice?

Always ask questions to guide learning. Never give direct answers."""
    
    # Build conversation with recent context
    conversation = f"System: {system_prompt}\n\n"
    
    # Add last 6 messages (3 pairs) for context but prioritize mode compliance
    recent_messages = messages[-6:] if len(messages) > 6 else messages
    
    for msg in recent_messages:
        role = msg.get("role", "")
        content = msg.get("content", "")
        if role == "user":
            conversation += f"Student: {content}\n"
        elif role == "assistant":
            conversation += f"Assistant: {content}\n"
    
    conversation += "Assistant:"
    return conversation

def validate_response_mode(response: str, is_force_mode: bool) -> str:
    """
    CRITICAL: Enforce mode compliance in responses
    """
    response = response.strip()
    
    if is_force_mode:
        # Force mode: Must be direct, no questions
        has_questioning = any(phrase in response.lower() for phrase in [
            "what do you think", "can you tell me", "what would happen", 
            "try it", "guess", "what pattern", "how do you", "what's your"
        ])
        
        if has_questioning or response.count("?") > 1:
            # Convert to direct answer
            print("🔧 Converting to direct answer for force mode")
            direct_parts = []
            for sentence in response.split("."):
                if "?" not in sentence and len(sentence.strip()) > 10:
                    direct_parts.append(sentence.strip())
            
            if direct_parts:
                return ". ".join(direct_parts[:2]) + "."
            else:
                return "Here's the direct answer: " + response.split("?")[0].strip() + "."
    
    else:
        # Mentor mode: Must have questions and guidance
        has_questions = "?" in response
        has_guidance = any(phrase in response.lower() for phrase in [
            "what do you think", "can you", "try", "what would", "how do you", "what pattern"
        ])
        
        if not has_questions and not has_guidance:
            # Convert to guiding questions
            print("🔧 Adding guiding questions for mentor mode")
            return f"Interesting! {response} What do you think about this? Can you tell me what part makes most sense to you?"
    
    return response

def generate_response(messages: list, is_force_mode: bool = False, max_tokens: int = 200, temperature: float = 0.7) -> str:
    """
    Generate response using the AI model with STRONG mode enforcement
    """
    try:
        # Create conversation prompt with strong mode directives
        prompt = create_conversation_prompt(messages, is_force_mode)
        
        print(f"🎯 Generating {'FORCE' if is_force_mode else 'MENTOR'} response with FIXED logic")
        print(f"🔍 DEBUG: force_mode = {is_force_mode}")
        print(f"📝 System prompt preview: {prompt.split('Student:')[0][:150]}...")
        
        # Adjust generation parameters based on mode
        if is_force_mode:
            # Force mode: Lower temperature for more focused, direct responses
            generation_temp = 0.2
            generation_tokens = min(max_tokens, 250)
        else:
            # Mentor mode: Slightly higher temperature for more varied questioning
            generation_temp = 0.4
            generation_tokens = min(max_tokens, 200)
        
        print(f"🎛️ Using temperature: {generation_temp}, max_tokens: {generation_tokens}")
        
        # Tokenize input
        inputs = tokenizer(prompt, return_tensors="pt", max_length=1024, truncation=True)
        
        # Generate response with mode-specific parameters
        with torch.no_grad():
            outputs = model.generate(
                inputs.input_ids,
                max_new_tokens=generation_tokens,
                temperature=generation_temp,
                do_sample=True,
                pad_token_id=tokenizer.eos_token_id,
                eos_token_id=tokenizer.eos_token_id,
                top_p=0.9,
                repetition_penalty=1.1
            )
        
        # Decode response
        full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Extract only the new generated part
        response = full_response[len(prompt):].strip()
        
        # Clean up response - remove role markers
        response = response.replace("Student:", "").replace("Assistant:", "").replace("System:", "").strip()
        
        # Remove any remaining conversation artifacts
        if "\n" in response:
            response = response.split("\n")[0].strip()
        
        print(f"✅ Raw generated response: {response[:100]}...")
        
        # CRITICAL: Validate and enforce mode compliance
        validated_response = validate_response_mode(response, is_force_mode)
        
        print(f"✅ Final validated response length: {len(validated_response)}")
        print(f"📝 Mode compliance: {'FORCE' if is_force_mode else 'MENTOR'}")
        
        if not validated_response or len(validated_response) < 10:
            # Strong fallback responses based on mode
            if is_force_mode:
                return "len() returns the number of items in a sequence. For example: len([1,2,3]) returns 3, len('hello') returns 5."
            else:
                return "What do you think len() might do? Try running len([1,2,3]) and see what happens! What number do you get?"
        
        return validated_response
        
    except Exception as e:
        print(f"❌ Generation error: {e}")
        # Mode-specific error fallbacks
        if is_force_mode:
            return "I need you to provide a more specific question so I can give you the exact answer you need."
        else:
            return "That's an interesting question! What do you think might be the answer? Can you break it down step by step?"

# === Routes ===
@app.get("/")
def root():
    return {
        "message": "🤖 Apollo AI Backend v3.1-FIXED - Qwen2-0.5B",
        "model": "Qwen/Qwen2-0.5B-Instruct with LoRA", 
        "status": "ready",
        "modes": {
            "mentor": "Guides learning with questions - FIXED",
            "force": "Provides direct answers - FIXED"
        },
        "fixes": "Strong mode enforcement, response validation"
    }

@app.get("/health")
def health():
    return {
        "status": "healthy", 
        "model_loaded": True, 
        "model_size": "0.5B",
        "version": "3.1-FIXED"
    }

@app.post("/v1/chat/completions")
async def chat_completions(request: Request):
    # Validate API key
    auth_header = request.headers.get("Authorization", "")
    if not auth_header.startswith("Bearer "):
        return JSONResponse(
            status_code=401, 
            content={"error": "Missing or invalid Authorization header"}
        )

    token = auth_header.replace("Bearer ", "").strip()
    if token != API_KEY:
        return JSONResponse(
            status_code=401, 
            content={"error": "Invalid API key"}
        )

    # Parse request body
    try:
        body = await request.json()
        messages = body.get("messages", [])
        max_tokens = min(body.get("max_tokens", 200), 400)
        temperature = max(0.1, min(body.get("temperature", 0.7), 1.0))
        
        # CRITICAL: Get force mode flag
        is_force_mode = body.get("force_mode", False)
        
        print(f"🚨 RECEIVED REQUEST - force_mode from body: {is_force_mode}")
        print(f"🚨 Type of force_mode: {type(is_force_mode)}")
        
        if not messages or not isinstance(messages, list):
            raise ValueError("Messages field is required and must be a list")
            
    except Exception as e:
        return JSONResponse(
            status_code=400, 
            content={"error": f"Invalid request body: {str(e)}"}
        )

    # Validate messages
    for i, msg in enumerate(messages):
        if not isinstance(msg, dict) or "role" not in msg or "content" not in msg:
            return JSONResponse(
                status_code=400,
                content={"error": f"Invalid message format at index {i}"}
            )

    try:
        print(f"📥 Processing request in {'FORCE' if is_force_mode else 'MENTOR'} mode - FIXED")
        print(f"📊 Total messages: {len(messages)}")
        print(f"🎯 CRITICAL - Mode flag received: {is_force_mode}")
        
        # Generate response with FIXED mode handling
        response_content = generate_response(
            messages=messages,
            is_force_mode=is_force_mode,
            max_tokens=max_tokens,
            temperature=temperature
        )
        
        print(f"✅ Generated response in {'FORCE' if is_force_mode else 'MENTOR'} mode")
        print(f"📝 Response preview: {response_content[:100]}...")
        
        return {
            "id": f"chatcmpl-apollo-{hash(str(messages)) % 10000}",
            "object": "chat.completion",
            "created": int(torch.tensor(0).item()),
            "model": f"qwen2-0.5b-{'force' if is_force_mode else 'mentor'}-fixed",
            "choices": [
                {
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "content": response_content
                    },
                    "finish_reason": "stop"
                }
            ],
            "usage": {
                "prompt_tokens": len(str(messages)),
                "completion_tokens": len(response_content),
                "total_tokens": len(str(messages)) + len(response_content)
            },
            "apollo_mode": "force" if is_force_mode else "mentor",
            "mode_validation": "FIXED - Strong enforcement",
            "model_optimizations": "qwen2_0.5B_fixed"
        }
        
    except Exception as e:
        print(f"❌ Chat completion error: {e}")
        return JSONResponse(
            status_code=500,
            content={"error": f"Internal server error: {str(e)}"}
        )

if __name__ == "__main__":
    import uvicorn
    print("🚀 Starting Apollo AI Backend v3.1-FIXED - Strong Mode Enforcement...")
    print("🧠 Model: Qwen/Qwen2-0.5B-Instruct (500M parameters)")
    print("🎯 Mentor Mode: FIXED - Always asks guiding questions")
    print("⚡ Force Mode: FIXED - Always gives direct answers")
    print("🔧 New: Response validation and mode enforcement")
    uvicorn.run(app, host="0.0.0.0", port=7860)