main.py

import os
import time
import random
import asyncio
import json
from fastapi import FastAPI, HTTPException, Depends
from fastapi.middleware.cors import CORSMiddleware
from fastapi.security.api_key import APIKeyHeader
from pydantic import BaseModel
from typing import List, Optional
from dotenv import load_dotenv
from starlette.responses import StreamingResponse
from openai import OpenAI
from typing import List, Optional, Dict, Any

load_dotenv()

BASE_URL = "https://generativelanguage.googleapis.com/v1beta/openai/"
EXPECTED_API_KEY = os.getenv("API_HUGGINGFACE")
API_KEY_NAME = "Authorization"
API_KEYS = [
    os.getenv("API_GEMINI_1"),
    os.getenv("API_GEMINI_2"),
    os.getenv("API_GEMINI_3"), 
    os.getenv("API_GEMINI_4"), 
    os.getenv("API_GEMINI_5") 
]

# Classi Pydantic di VALIDAZIONE Body
class Message(BaseModel):
    role: Any
    content: Any

class ChatCompletionRequest(BaseModel):
    model: str = "gemini-2.0-flash" 
    messages: List[Message]
    max_tokens: Optional[int] = 8196
    temperature: Optional[float] = 0.8
    stream: Optional[bool] = False
    stream_options: Optional[Dict[str, Any]] = None
    class Config:
        extra = "allow"

# Server FAST API 
app = FastAPI(title="OpenAI-SDK-compatible API", version="1.0.0", description="Un wrapper FastAPI compatibile con le specifiche dell'API OpenAI.")
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

# Validazione API       
api_key_header = APIKeyHeader(name=API_KEY_NAME, auto_error=False)
def verify_api_key(api_key: str = Depends(api_key_header)):
    ''' Validazione Chiave API - Per ora in ENV, Token HF '''
    if not api_key:
        raise HTTPException(status_code=403, detail="API key mancante")
    if api_key != f"Bearer {EXPECTED_API_KEY}":
        raise HTTPException(status_code=403, detail="API key non valida")
    return api_key

# Client OpenAI
def get_openai_client():
    ''' Client OpenAI passando in modo RANDOM le Chiavi API. In questo modo posso aggirare i limiti "Quota Exceeded" '''
    api_key = random.choice(API_KEYS)
    return OpenAI(api_key=api_key, base_url=BASE_URL)

# Chiama API (senza Streaming)
def call_api_sync(params: ChatCompletionRequest):
    ''' Chiamata API senza streaming. Se da errore 429 lo rifa'''
    try:
        client = get_openai_client()
        response_format = getattr(params, 'response_format', None)
        if response_format and getattr(response_format, 'type', None) == 'json_schema':    
            response = client.beta.chat.completions.parse(**params.model_dump())
        else:
            response = client.chat.completions.create(**params.model_dump())
        return response
    except Exception as e:
        if "429" in str(e):
            time.sleep(2)
            return call_api_sync(params)
        else:
            raise e

# Chiama API (con Streaming)
async def _resp_async_generator(params: ChatCompletionRequest):
    ''' Chiamata API con streaming. Se da errore 429 lo rifa'''
    client = get_openai_client()
    try:
        response = client.chat.completions.create(**params.model_dump())
        for chunk in response:
            chunk_data = chunk.to_dict() if hasattr(chunk, "to_dict") else chunk
            yield f"data: {json.dumps(chunk_data)}\n\n"
            await asyncio.sleep(0.01)
        yield "data: [DONE]\n\n"
    except Exception as e:
        if "429" in str(e):
            await asyncio.sleep(2)
            async for item in _resp_async_generator(params):
                yield item
        else:
            error_data = {"error": str(e)}
            yield f"data: {json.dumps(error_data)}\n\n"

# ---------------------------------- Metodi API ---------------------------------------
@app.get("/")
def read_general(): 
    return {"response": "Benvenuto"}

@app.get("/health")
async def health_check():
    return {"message": "success"}

@app.post("/v1/chat/completions", dependencies=[Depends(verify_api_key)])
async def chat_completions(req: ChatCompletionRequest):
    print(req)
    try:    
        if not req.messages:
            raise HTTPException(status_code=400, detail="Nessun messaggio fornito")
        if not req.stream:
            return call_api_sync(req)
        else:
            return StreamingResponse(_resp_async_generator(req), media_type="application/x-ndjson")            
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))