app.py

import os
import sqlite3
import logging
import anyio
import uvicorn
from fastapi import FastAPI, Depends, HTTPException, Security
from fastapi.security import HTTPAuthorizationCredentials, HTTPBearer
from fastapi.middleware.cors import CORSMiddleware
from fastapi.responses import StreamingResponse
from pydantic import BaseModel
from llama_cpp import Llama

# --- INITIALIZATION ---
app = FastAPI(title="Kwen Foundation Official API")

# Updated CORS: allow_methods=["*"] is crucial for browser 'OPTIONS' preflight requests
app.add_middleware(
    CORSMiddleware,
    allow_origins=["https://hydrogenclient.github.io"], # For debugging, allowing all. Change to your specific domain later.
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

logging.basicConfig(level=logging.INFO, format="%(message)s")
logger = logging.getLogger(__name__)
security = HTTPBearer()

BASE_DIR = os.path.dirname(os.path.abspath(__file__))
MODEL_PATH = os.path.join(BASE_DIR, "kwen-4b-q4_k_m.gguf")
DB_PATH = "keys.db"

# --- MODEL LOADING ---
if os.path.exists(MODEL_PATH):
    size_gb = os.path.getsize(MODEL_PATH) / (1024**3)
    logger.info(f"✅ Model found! Size: {size_gb:.2f} GB")
    llm = Llama(
        model_path=MODEL_PATH,
        n_ctx=2048,
        n_threads=2,
        n_gpu_layers=0,
        offload_kqv=False,
        verbose=True
    )
    logger.info("Kwen Model loaded.")
else:
    logger.error(f"❌ Model NOT found at {MODEL_PATH}")
    llm = None

# --- AUTH SETUP ---
DEFAULT_KEY = "kwen_NjdT7VQauEyDfzVvPLMIQfwUuduJWT6j"

def init_db():
    conn = sqlite3.connect(DB_PATH)
    conn.execute("CREATE TABLE IF NOT EXISTS keys (key TEXT PRIMARY KEY, user TEXT)")
    cur = conn.execute("SELECT user FROM keys WHERE key = ?", (DEFAULT_KEY,))
    if not cur.fetchone():
        conn.execute("INSERT INTO keys (key, user) VALUES (?, ?)", (DEFAULT_KEY, "Public_Guest"))
        conn.commit()
    conn.close()

init_db()

class ChatInput(BaseModel):
    prompt: str
    max_tokens: int = 512
    temperature: float = 0.6

def verify_key(auth: HTTPAuthorizationCredentials = Security(security)):
    provided_key = auth.credentials.strip()
    conn = sqlite3.connect(DB_PATH)
    user = conn.execute("SELECT user FROM keys WHERE key = ?", (provided_key,)).fetchone()
    conn.close()
    
    if not user:
        logger.warning(f"AUTH FAILED: Access denied for key ending in ...{provided_key[-4:]}")
        raise HTTPException(status_code=403, detail="Invalid API Key")
    return user[0]

# --- STREAMING ENGINE ---
async def kwen_streamer(prompt, max_tokens, temperature):
    formatted = f"<|im_start|>user\n{prompt}<|im_end|>\n<|im_start|>assistant\n"
    yield "" # Initial heartbeat

    def run_inference():
        return llm(
            formatted,
            max_tokens=max_tokens,
            temperature=temperature,
            stop=["<|im_end|>", "user:", "assistant:"],
            stream=True
        )

    try:
        stream = await anyio.to_thread.run_sync(run_inference)
        for chunk in stream:
            token = chunk["choices"][0]["text"]
            if token:
                yield token
            await anyio.sleep(0.01)
    except Exception as e:
        logger.error(f"Stream Error: {e}")
        yield f"\n[Stream Error: {e}]"

# --- ENDPOINTS ---
@app.get("/")
def health():
    return {"status": "running", "model": "Kwen-4B", "database": "active"}

@app.post("/v1/chat")
async def chat(data: ChatInput, user=Depends(verify_key)):
    if not llm:
        raise HTTPException(status_code=503, detail="Model Offline")
    logger.info(f"Authorized Request from: {user}")
    return StreamingResponse(
        kwen_streamer(data.prompt, data.max_tokens, data.temperature),
        media_type="text/event-stream"
    )

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