app.py

from fastapi import FastAPI, Request, HTTPException, Header
from fastapi.responses import JSONResponse, StreamingResponse
from fastapi.middleware.cors import CORSMiddleware
from typing import Optional, List, Dict, Any
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch
import time
import os
import json
import secrets
from datetime import datetime, timedelta

# Indian Standard Time helper
def get_ist_time():
    """Get current time in IST (UTC+5:30)"""
    return datetime.utcnow() + timedelta(hours=5, minutes=30)

# Load model directly in the Space (FREE and UNLIMITED!)
# AJ-Mini: Unlimited free assistant by AJ STUDIOZ  
MODEL_NAME = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
BRANDING_NAME = "AJ-Mini v1.0"
DEVELOPER = "AJ STUDIOZ"

# Initialize model and tokenizer with optimization
print(f"Loading {BRANDING_NAME} (based on {MODEL_NAME})...")
print(f"Developed by {DEVELOPER}")
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(
    MODEL_NAME,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    low_cpu_mem_usage=True,
    device_map="auto"
)
model.eval()  # Set to evaluation mode
print(f"{BRANDING_NAME} loaded successfully!")

def query_ollama_model(prompt: str, max_tokens: int = 1000, temperature: float = 0.7, stream: bool = False):
    """Query model loaded directly in the Space - Optimized for speed"""
    try:
        # Optimize for faster responses
        max_tokens = min(max_tokens, 150)  # Reduced from 256 to 150 for faster CPU inference
        
        # Tokenize input
        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512).to(model.device)
        
        # Generate response with optimization
        with torch.no_grad():  # Disable gradient computation for faster inference
            outputs = model.generate(
                **inputs,
                max_new_tokens=max_tokens,
                temperature=temperature,
                do_sample=temperature > 0.1,
                top_p=0.9,
                top_k=40,  # Reduced from 50 to 40 for speed
                repetition_penalty=1.15,  # Reduce repetition
                pad_token_id=tokenizer.eos_token_id,
                eos_token_id=tokenizer.eos_token_id,
                num_beams=1,  # Greedy decoding for speed
                early_stopping=True,
                no_repeat_ngram_size=3  # Prevent repetition
            )
        
        # Extract only the generated text (remove input)
        input_length = inputs.input_ids.shape[1]
        generated_tokens = outputs[0][input_length:]
        generated_text = tokenizer.decode(generated_tokens, skip_special_tokens=True).strip()
        
        # Create response object
        class SuccessResponse:
            status_code = 200
            def json(self):
                return [{"generated_text": generated_text}]
            text = generated_text
        
        return SuccessResponse()
    except Exception as e:
        # Create error response
        class ErrorResponse:
            status_code = 500
            def json(self):
                return {"error": str(e)}
            text = str(e)
        return ErrorResponse()

# Simple API key validation for AJ format
VALID_API_KEY_PREFIX = "aj_"

# Anthropic API key validation
def validate_anthropic_key(api_key: Optional[str]) -> bool:
    """Validate Anthropic-style API key"""
    if not api_key:
        return False
    return api_key.startswith("sk-ant-") and len(api_key) > 20

def validate_api_key(api_key: Optional[str]) -> bool:
    """Validate API key format - accepts both AJ and Anthropic formats"""
    if not api_key:
        return False
    return (api_key.startswith(VALID_API_KEY_PREFIX) and len(api_key) > 10) or validate_anthropic_key(api_key)

def extract_api_key(authorization: Optional[str]) -> Optional[str]:
    """Extract API key from Authorization header"""
    if not authorization:
        return None
    if authorization.startswith("Bearer "):
        return authorization[7:]
    return authorization

def extract_anthropic_key(x_api_key: Optional[str]) -> Optional[str]:
    """Extract API key from x-api-key header (Anthropic style)"""
    return x_api_key

app = FastAPI(
    title="AJ STUDIOZ Mini API",
    version="1.0",
    description="AJ-Mini: Unlimited free AI assistant by AJ STUDIOZ - Powered by TinyLlama-1.1B"
)

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

@app.get("/")
async def root():
    return {
        "service": "AJ STUDIOZ Mini API",
        "version": "1.0",
        "model": "AJ-Mini v1.0 (TinyLlama-1.1B-Chat)",
        "developed_by": "AJ STUDIOZ",
        "status": "online",
        "provider": "AJ STUDIOZ",
        "website": "https://ajstudioz.co.in",
        "pricing": {
            "plan": "LIFETIME FREE",
            "rate_limits": "UNLIMITED",
            "cost": "FREE FOREVER",
            "usage_cap": "NONE"
        },
        "description": "AJ-Mini: Unlimited free AI assistant by AJ STUDIOZ with Claude API compatibility, OpenAI support, and reliable 24/7 availability",
        "capabilities": [
            "General conversation & assistance",
            "Basic code generation & help",
            "Multi-language support",
            "Simple problem solving",
            "Anthropic Claude API compatible",
            "OpenAI-compatible API",
            "Text generation & responses",
            "Markdown formatting",
            "24/7 availability",
            "Enterprise security",
            "Unlimited usage - FREE FOREVER"
        ],
        "endpoints": {
            "v1_messages": "/v1/messages - Anthropic Claude-compatible endpoint",
            "v1_chat": "/v1/chat/completions - OpenAI-compatible chat endpoint",
            "v1_completions": "/v1/completions - OpenAI-compatible completions",
            "v1_models": "/v1/models - List available models",
            "chat": "/chat - Simple chat interface",
            "generate": "/api/generate - Direct generation API"
        },
        "authentication": {
            "anthropic": "x-api-key: sk-ant-<your_key>",
            "openai": "Authorization: Bearer aj_<your_key>",
            "note": "Both formats accepted for compatibility"
        }
    }

@app.post("/v1/messages")
async def anthropic_messages(
    request: Request,
    x_api_key: Optional[str] = Header(None, alias="x-api-key"),
    anthropic_version: Optional[str] = Header(None, alias="anthropic-version")
):
    """Anthropic Claude-compatible messages endpoint"""
    # Validate API key
    api_key = extract_anthropic_key(x_api_key)
    if not validate_api_key(api_key):
        return JSONResponse(
            status_code=401,
            content={
                "type": "error",
                "error": {
                    "type": "authentication_error",
                    "message": "Invalid API key. Use format: sk-ant-<your_key> or aj_<your_key>"
                }
            }
        )
    
    try:
        data = await request.json()
        messages = data.get("messages", [])
        model = data.get("model", "claude-sonnet-4-20250514")
        max_tokens = data.get("max_tokens", 1024)
        temperature = data.get("temperature", 1.0)
        stream = data.get("stream", False)
        
        if not messages:
            return JSONResponse(
                status_code=400,
                content={
                    "type": "error",
                    "error": {
                        "type": "invalid_request_error",
                        "message": "messages is required"
                    }
                }
            )
        
        # Convert to prompt format for text_generation
        prompt_parts = ["You are AJ Mini, a helpful AI assistant created by AJ STUDIOZ. You provide reliable assistance and are available 24/7 unlimited and free.\n"]
        
        for msg in messages:
            role = msg.get("role")
            content = msg.get("content")
            if isinstance(content, list):
                # Handle complex content (text, images, etc.)
                text_parts = [c.get("text", "") for c in content if c.get("type") == "text"]
                content = " ".join(text_parts)
            
            if role == "user":
                prompt_parts.append(f"User: {content}")
            elif role == "assistant":
                prompt_parts.append(f"Assistant: {content}")
            elif role == "system":
                prompt_parts.insert(0, content)
        
        prompt_parts.append("Assistant:")
        full_prompt = "\n\n".join(prompt_parts)
        
        response = query_ollama_model(full_prompt, max_tokens, temperature)
        
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                assistant_message = result[0].get('generated_text', '')
            else:
                assistant_message = result.get('generated_text', '')
        else:
            raise HTTPException(status_code=500, detail=f"Model error: {response.text}")
        
        # Return Anthropic-compatible response
        return {
            "id": f"msg_{secrets.token_hex(12)}",
            "type": "message",
            "role": "assistant",
            "content": [
                {
                    "type": "text",
                    "text": assistant_message
                }
            ],
            "model": model,
            "stop_reason": "end_turn",
            "stop_sequence": None,
            "usage": {
                "input_tokens": sum(len(str(m.get("content", "")).split()) for m in messages),
                "output_tokens": len(assistant_message.split())
            }
        }
        
    except HTTPException:
        raise
    except Exception as e:
        return JSONResponse(
            status_code=500,
            content={
                "type": "error",
                "error": {
                    "type": "api_error",
                    "message": str(e)
                }
            }
        )

@app.get("/v1/models")
async def list_models(authorization: Optional[str] = Header(None)):
    """OpenAI-compatible models endpoint"""
    api_key = extract_api_key(authorization)
    if not validate_api_key(api_key):
        raise HTTPException(status_code=401, detail="Invalid API key. Use format: aj_your_key")
    
    return {
        "object": "list",
        "data": [
            {
                "id": "aj-mini",
                "object": "model",
                "created": 1730505600,
                "owned_by": "aj-studioz",
                "permission": [],
                "root": "aj-mini",
                "parent": None,
            },
            {
                "id": "aj-mini-v1",
                "object": "model",
                "created": 1730505600,
                "owned_by": "aj-studioz",
                "permission": [],
                "root": "aj-mini-v1",
                "parent": None,
            }
        ]
    }

@app.post("/v1/chat/completions")
async def chat_completions(request: Request, authorization: Optional[str] = Header(None)):
    """OpenAI-compatible chat completions endpoint - Public access for AJ STUDIOZ"""
    # Make API key optional for public access
    # api_key = extract_api_key(authorization)
    # if not validate_api_key(api_key):
    #     raise HTTPException(
    #         status_code=401, 
    #         detail={
    #             "error": {
    #                 "message": "Invalid API key. Your API key should start with 'aj_'",
    #                 "type": "invalid_request_error",
    #                 "code": "invalid_api_key"
    #             }
    #         }
    #     )
    
    try:
        data = await request.json()
        messages = data.get("messages", [])
        model = data.get("model", "aj-mini")
        max_tokens = data.get("max_tokens", 256)
        temperature = data.get("temperature", 0.7)
        
        if not messages:
            raise HTTPException(status_code=400, detail="Messages are required")
        
        # Add system prompt if not provided
        has_system = any(msg.get("role") == "system" for msg in messages)
        if not has_system:
            messages.insert(0, {
                "role": "system",
                "content": "You are AJ-Mini v1.0, a helpful AI assistant developed by AJ STUDIOZ. You are friendly, concise, and helpful. When asked who you are, introduce yourself as AJ-Mini created by AJ STUDIOZ."
            })
        
        # Convert messages to prompt
        prompt_parts = []
        for msg in messages:
            role = msg.get("role", "user")
            content = msg.get("content", "")
            if role == "system":
                prompt_parts.append(f"{content}")
            elif role == "user":
                prompt_parts.append(f"User: {content}")
            elif role == "assistant":
                prompt_parts.append(f"Assistant: {content}")
        
        prompt = "\n\n".join(prompt_parts) + "\n\nAssistant:"
        completion_id = f"chatcmpl-{secrets.token_hex(12)}"
        
        response = query_ollama_model(prompt, max_tokens, temperature)
        
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                assistant_message = result[0].get('generated_text', '')
            else:
                assistant_message = result.get('generated_text', '')
        else:
            raise HTTPException(status_code=500, detail=f"Model error: {response.text}")
        
        # OpenAI-compatible response
        return {
            "id": completion_id,
            "object": "chat.completion",
            "created": int(time.time()),
            "model": model,
            "choices": [
                {
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "content": assistant_message
                    },
                    "finish_reason": "stop"
                }
            ],
            "usage": {
                "prompt_tokens": len(prompt.split()),
                "completion_tokens": len(assistant_message.split()),
                "total_tokens": len(prompt.split()) + len(assistant_message.split())
            },
            "system_fingerprint": "aj-mini-v1.0"
        }
            
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/v1/completions")
async def completions(request: Request, authorization: Optional[str] = Header(None)):
    """OpenAI-compatible completions endpoint"""
    api_key = extract_api_key(authorization)
    if not validate_api_key(api_key):
        raise HTTPException(status_code=401, detail="Invalid API key")
    
    try:
        data = await request.json()
        prompt = data.get("prompt", "")
        model = data.get("model", "aj-mini")
        max_tokens = data.get("max_tokens", 2000)
        temperature = data.get("temperature", 0.7)
        
        if not prompt:
            raise HTTPException(status_code=400, detail="Prompt is required")
        
        response = query_ollama_model(prompt, max_tokens, temperature)
        
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                completion_text = result[0].get('generated_text', '')
            else:
                completion_text = result.get('generated_text', '')
        else:
            raise HTTPException(status_code=500, detail=f"Model error: {response.text}")
        
        return {
            "id": f"cmpl-{secrets.token_hex(12)}",
            "object": "text_completion",
            "created": int(time.time()),
            "model": model,
            "choices": [
                {
                    "text": completion_text,
                    "index": 0,
                    "logprobs": None,
                    "finish_reason": "stop"
                }
            ],
            "usage": {
                "prompt_tokens": len(prompt.split()),
                "completion_tokens": len(completion_text.split()),
                "total_tokens": len(prompt.split()) + len(completion_text.split())
            }
        }
            
    except HTTPException:
        raise
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.post("/chat")
async def chat(request: Request):
    try:
        data = await request.json()
        message = data.get("message", "")
        
        if not message:
            return JSONResponse({"error": "Message is required"}, status_code=400)
        
        # Simple prompt for Qwen
        full_message = f"You are AJ, a helpful AI assistant by AJ STUDIOZ.\n\nUser: {message}\n\nAssistant:"
        
        response = query_ollama_model(full_message, 500, 0.7)
        
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                reply = result[0].get('generated_text', '')
            else:
                reply = result.get('generated_text', '')
            
            return JSONResponse({
                "reply": reply,
                "model": "AJ-Mini v1.0",
                "provider": "AJ STUDIOZ"
            })
        else:
            return JSONResponse(
                {"error": "Model error", "details": response.text},
                status_code=500
            )
            
    except Exception as e:
        return JSONResponse(
            {"error": "Failed to process request", "details": str(e)},
            status_code=500
        )

@app.post("/api/generate")
async def generate(request: Request):
    """Direct API for text generation"""
    try:
        data = await request.json()
        prompt = data.get("prompt", "")
        max_tokens = data.get("max_tokens", 1000)
        temperature = data.get("temperature", 0.7)
        
        if not prompt:
            return JSONResponse({"error": "Prompt is required"}, status_code=400)
        
        response = query_ollama_model(prompt, max_tokens, temperature)
        
        if response.status_code == 200:
            result = response.json()
            if isinstance(result, list) and len(result) > 0:
                response_text = result[0].get('generated_text', '')
            else:
                response_text = result.get('generated_text', '')
            
            return JSONResponse({
                "response": response_text,
                "model": "AJ-Mini v1.0",
                "done": True
            })
        else:
            return JSONResponse(
                {"error": "Model error", "details": response.text},
                status_code=500
            )
        
    except Exception as e:
        return JSONResponse(
            {"error": "Failed to process request", "details": str(e)},
            status_code=500
        )

@app.get("/health")
async def health():
    """Fast health check endpoint - no model query"""
    return {
        "status": "healthy",
        "service": "AJ STUDIOZ Mini API",
        "model": "AJ-Mini v1.0",
        "version": "1.0",
        "developer": "AJ STUDIOZ",
        "platform": "HuggingFace Spaces (CPU)",
        "availability": "Unlimited FREE",
        "timestamp": get_ist_time().strftime("%Y-%m-%d %H:%M:%S IST"),
        "note": "Use POST /v1/chat/completions for inference"
    }

@app.get("/test")
@app.get("/ping")
async def quick_test():
    """Ultra-fast test endpoint for ReqBin - responds in < 200ms"""
    return {
        "status": "ok",
        "message": "AJ-Mini v1.0 is operational",
        "model": "aj-mini",
        "latency": "< 200ms",
        "endpoint": "POST /v1/chat/completions",
        "example": {
            "model": "aj-mini",
            "messages": [{"role": "user", "content": "Hello"}]
        },
        "developer": "AJ STUDIOZ",
        "availability": "UNLIMITED FREE",
        "timestamp": get_ist_time().strftime("%Y-%m-%d %H:%M:%S IST")
    }

if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)