app.py

import os
import time
import logging
import requests
import json
import uuid
import concurrent.futures
import threading
import base64
import io
from PIL import Image
from datetime import datetime
from apscheduler.schedulers.background import BackgroundScheduler
from flask import Flask, request, jsonify, Response, stream_with_context

os.environ['TZ'] = 'Asia/Shanghai'
time.tzset()

logging.basicConfig(level=logging.INFO,
                    format='%(asctime)s - %(levelname)s - %(message)s')

API_ENDPOINT = "https://api-st.siliconflow.cn/v1/user/info"
TEST_MODEL_ENDPOINT = "https://api-st.siliconflow.cn/v1/chat/completions"
MODELS_ENDPOINT = "https://api-st.siliconflow.cn/v1/models"
EMBEDDINGS_ENDPOINT = "https://api-st.siliconflow.cn/v1/embeddings"

app = Flask(__name__)

text_models = []
free_text_models = []
embedding_models = []
free_embedding_models = []
image_models = []
free_image_models = []

invalid_keys_global = []
free_keys_global = []
unverified_keys_global = []
valid_keys_global = []

executor = concurrent.futures.ThreadPoolExecutor(max_workers=10000)
model_key_indices = {}

request_timestamps = []
token_counts = []
request_timestamps_day = []
token_counts_day = []
data_lock = threading.Lock()

def get_credit_summary(api_key):
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    max_retries = 3

    for attempt in range(max_retries):
        try:
            response = requests.get(API_ENDPOINT, headers=headers, timeout=2)
            response.raise_for_status()
            data = response.json().get("data", {})
            total_balance = data.get("totalBalance", 0)
            logging.info(f"获取额度，API Key：{api_key}，当前额度: {total_balance}")
            return {"total_balance": float(total_balance)}
        except requests.exceptions.Timeout as e:
            logging.error(f"获取额度信息失败，API Key：{api_key}，尝试次数：{attempt+1}/{max_retries}，错误信息：{e} (Timeout)")
            if attempt < max_retries - 1:
                pass
            else:
                logging.error(f"获取额度信息失败，API Key：{api_key}，所有重试次数均已失败 (Timeout)")
        except requests.exceptions.RequestException as e:
            logging.error(f"获取额度信息失败，API Key：{api_key}，错误信息：{e}")
            return None
    return None

FREE_MODEL_TEST_KEY = (
    "sk-bmjbjzleaqfgtqfzmcnsbagxrlohriadnxqrzfocbizaxukw"
)

FREE_IMAGE_LIST = [
    "stabilityai/stable-diffusion-3-5-large",
    "black-forest-labs/FLUX.1-schnell",
    "stabilityai/stable-diffusion-3-medium",
    "stabilityai/stable-diffusion-xl-base-1.0",
    "stabilityai/stable-diffusion-2-1"
]

def test_model_availability(api_key, model_name):
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    try:
        response = requests.post(
            TEST_MODEL_ENDPOINT,
            headers=headers,
            json={
                "model": model_name,
                "messages": [{"role": "user", "content": "hi"}],
                "max_tokens": 5,
                "stream": False
            },
            timeout=5
        )
        if response.status_code == 429 or response.status_code == 200:
            return True
        else:
            return False
    except requests.exceptions.RequestException as e:
        logging.error(
            f"测试模型 {model_name} 可用性失败，"
            f"API Key：{api_key}，错误信息：{e}"
        )
        return False

def refresh_models():
    global text_models, free_text_models
    global embedding_models, free_embedding_models
    global image_models, free_image_models

    text_models = get_all_models(FREE_MODEL_TEST_KEY, "chat")
    embedding_models = get_all_models(FREE_MODEL_TEST_KEY, "embedding")
    image_models = get_all_models(FREE_MODEL_TEST_KEY, "text-to-image")
    free_text_models = []
    free_embedding_models = []
    free_image_models = []

    ban_models_str = os.environ.get("BAN_MODELS")
    ban_models = []
    if ban_models_str:
        try:
            ban_models = json.loads(ban_models_str)
            if not isinstance(ban_models, list):
                logging.warning(
                    "环境变量 BAN_MODELS 格式不正确，应为 JSON 数组。"
                )
                ban_models = []
        except json.JSONDecodeError:
            logging.warning(
                "环境变量 BAN_MODELS JSON 解析失败，请检查格式。"
            )
            ban_models = []
    
    text_models = [model for model in text_models if model not in ban_models]
    embedding_models = [model for model in embedding_models if model not in ban_models]
    image_models = [model for model in image_models if model not in ban_models]

    with concurrent.futures.ThreadPoolExecutor(
        max_workers=10000
    ) as executor:
        future_to_model = {
            executor.submit(
                test_model_availability,
                FREE_MODEL_TEST_KEY,
                model
            ): model for model in text_models
        }
        for future in concurrent.futures.as_completed(future_to_model):
            model = future_to_model[future]
            try:
                is_free = future.result()
                if is_free:
                    free_text_models.append(model)
            except Exception as exc:
                logging.error(f"模型 {model} 测试生成异常: {exc}")

    with concurrent.futures.ThreadPoolExecutor(
        max_workers=10000
    ) as executor:
        future_to_model = {
            executor.submit(
                test_embedding_model_availability,
                FREE_MODEL_TEST_KEY, model
            ): model for model in embedding_models
        }
        for future in concurrent.futures.as_completed(future_to_model):
            model = future_to_model[future]
            try:
                is_free = future.result()
                if is_free:
                    free_embedding_models.append(model)
            except Exception as exc:
                logging.error(f"模型 {model} 测试生成异常: {exc}")
                
    with concurrent.futures.ThreadPoolExecutor(
        max_workers=10000
    ) as executor:
        future_to_model = {
            executor.submit(
                test_image_model_availability,
                FREE_MODEL_TEST_KEY, model
            ): model for model in image_models
        }
        for future in concurrent.futures.as_completed(future_to_model):
            model = future_to_model[future]
            try:
                is_free = future.result()
                if is_free:
                    free_image_models.append(model)
            except Exception as exc:
                logging.error(f"模型 {model} 测试生成异常: {exc}")

    logging.info(f"所有文本模型列表：{text_models}")
    logging.info(f"免费文本模型列表：{free_text_models}")
    logging.info(f"所有向量模型列表：{embedding_models}")
    logging.info(f"免费向量模型列表：{free_embedding_models}")
    logging.info(f"所有生图模型列表：{image_models}")
    logging.info(f"免费生图模型列表：{free_image_models}")

def test_embedding_model_availability(api_key, model_name):
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    try:
        response = requests.post(
            EMBEDDINGS_ENDPOINT,
            headers=headers,
            json={
                "model": model_name,
                "input": ["hi"],
            },
            timeout=10
        )
        if response.status_code == 429 or response.status_code == 200:
            return True
        else:
            return False
    except requests.exceptions.RequestException as e:
        logging.error(
            f"测试向量模型 {model_name} 可用性失败，"
            f"API Key：{api_key}，错误信息：{e}"
        )
        return False
        
def test_image_model_availability(api_key, model_name):
    return model_name in FREE_IMAGE_LIST

def load_keys():
    keys_str = os.environ.get("KEYS")
    test_model = os.environ.get(
        "TEST_MODEL",
        "Pro/google/gemma-2-9b-it"
    )

    if keys_str:
        keys = [key.strip() for key in keys_str.split(',')]
        unique_keys = list(set(keys))
        keys_str = ','.join(unique_keys) 
        os.environ["KEYS"] = keys_str

        logging.info(f"加载的 keys：{unique_keys}")

        with concurrent.futures.ThreadPoolExecutor(
            max_workers=10000
        ) as executor:
            future_to_key = {
                executor.submit(
                    process_key, key, test_model
                ): key for key in unique_keys
            }

            invalid_keys = []
            free_keys = []
            unverified_keys = []
            valid_keys = []

            for future in concurrent.futures.as_completed(
                future_to_key
            ):
                key = future_to_key[future]
                try:
                    key_type = future.result()
                    if key_type == "invalid":
                        invalid_keys.append(key)
                    elif key_type == "free":
                        free_keys.append(key)
                    elif key_type == "unverified":
                        unverified_keys.append(key)
                    elif key_type == "valid":
                        valid_keys.append(key)
                except Exception as exc:
                    logging.error(f"处理 KEY {key} 生成异常: {exc}")

        logging.info(f"无效 KEY：{invalid_keys}")
        logging.info(f"免费 KEY：{free_keys}")
        logging.info(f"未实名 KEY：{unverified_keys}")
        logging.info(f"有效 KEY：{valid_keys}")

        global invalid_keys_global, free_keys_global
        global unverified_keys_global, valid_keys_global
        invalid_keys_global = invalid_keys
        free_keys_global = free_keys
        unverified_keys_global = unverified_keys
        valid_keys_global = valid_keys

    else:
        logging.warning("环境变量 KEYS 未设置。")

def process_key(key, test_model):
    credit_summary = get_credit_summary(key)
    if credit_summary is None:
        return "invalid"
    else:
        total_balance = credit_summary.get("total_balance", 0)
        if total_balance <= 0.03:
            return "free"
        else:
            if test_model_availability(key, test_model):
                return "valid"
            else:
                return "unverified"

def get_all_models(api_key, sub_type):
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    try:
        response = requests.get(
            MODELS_ENDPOINT,
            headers=headers,
            params={"sub_type": sub_type}
        )
        response.raise_for_status()
        data = response.json()
        if (
            isinstance(data, dict) and
            'data' in data and
            isinstance(data['data'], list)
        ):
            return [
                model.get("id") for model in data["data"]
                if isinstance(model, dict) and "id" in model
            ]
        else:
            logging.error("获取模型列表失败：响应数据格式不正确")
            return []
    except requests.exceptions.RequestException as e:
        logging.error(
            f"获取模型列表失败，"
            f"API Key：{api_key}，错误信息：{e}"
        )
        return []
    except (KeyError, TypeError) as e:
        logging.error(
            f"解析模型列表失败，"
            f"API Key：{api_key}，错误信息：{e}"
        )
        return []

def determine_request_type(model_name, model_list, free_model_list):
    if model_name in free_model_list:
        return "free"
    elif model_name in model_list:
        return "paid"
    else:
        return "unknown"

def select_key(request_type, model_name):
    if request_type == "free":
        available_keys = (
            free_keys_global +
            unverified_keys_global +
            valid_keys_global
        )
    elif request_type == "paid":
        available_keys = unverified_keys_global + valid_keys_global
    else:
        available_keys = (
            free_keys_global +
            unverified_keys_global +
            valid_keys_global
        )

    if not available_keys:
        return None

    current_index = model_key_indices.get(model_name, 0)

    for _ in range(len(available_keys)):
        key = available_keys[current_index % len(available_keys)]
        current_index += 1

        if key_is_valid(key, request_type):
            model_key_indices[model_name] = current_index
            return key
        else:
            logging.warning(
                f"KEY {key} 无效或达到限制，尝试下一个 KEY"
            )

    model_key_indices[model_name] = 0
    return None

def key_is_valid(key, request_type):
    if request_type == "invalid":
        return False

    credit_summary = get_credit_summary(key)
    if credit_summary is None:
        return False

    total_balance = credit_summary.get("total_balance", 0)

    if request_type == "free":
        return True
    elif request_type == "paid" or request_type == "unverified":
        return total_balance > 0
    else:
        return False

def check_authorization(request):
    authorization_key = os.environ.get("AUTHORIZATION_KEY")
    if not authorization_key:
        logging.warning("环境变量 AUTHORIZATION_KEY 未设置，请设置后重试。")
        return False

    auth_header = request.headers.get('Authorization')
    if not auth_header:
        logging.warning("请求头中缺少 Authorization 字段。")
        return False

    if auth_header != f"Bearer {authorization_key}":
        logging.warning(f"无效的 Authorization 密钥：{auth_header}")
        return False

    return True

scheduler = BackgroundScheduler()
scheduler.add_job(load_keys, 'interval', hours=1)
scheduler.remove_all_jobs()
scheduler.add_job(refresh_models, 'interval', hours=1)

@app.route('/')
def index():
    current_time = time.time()
    one_minute_ago = current_time - 60
    one_day_ago = current_time - 86400

    with data_lock:
        while request_timestamps and request_timestamps[0] < one_minute_ago:
            request_timestamps.pop(0)
            token_counts.pop(0)

        rpm = len(request_timestamps)
        tpm = sum(token_counts)

    with data_lock:
        while request_timestamps_day and request_timestamps_day[0] < one_day_ago:
            request_timestamps_day.pop(0)
            token_counts_day.pop(0)

        rpd = len(request_timestamps_day)
        tpd = sum(token_counts)

    return jsonify({"rpm": rpm, "tpm": tpm, "rpd": rpd, "tpd": tpd})
            
@app.route('/handsome/v1/models', methods=['GET'])
def list_models():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    detailed_models = []
    
    for model in text_models:
        detailed_models.append({
            "id": model,
            "object": "model",
            "created": 1678888888,
            "owned_by": "openai",
            "permission": [
            ],
            "root": model,
            "parent": None
        })

    for model in embedding_models:
        detailed_models.append({
            "id": model,
            "object": "model",
            "created": 1678888888,
            "owned_by": "openai",
            "permission": [
            ],
            "root": model,
            "parent": None
        })
    
    for model in image_models:
        detailed_models.append({
            "id": model,
            "object": "model",
            "created": 1678888888,
            "owned_by": "openai",
            "permission": [
            ],
            "root": model,
            "parent": None
        })

    return jsonify({
        "success": True,
        "data": detailed_models
    })

def get_billing_info():
    keys = valid_keys_global + unverified_keys_global
    total_balance = 0

    with concurrent.futures.ThreadPoolExecutor(
        max_workers=10000
    ) as executor:
        futures = [
            executor.submit(get_credit_summary, key) for key in keys
        ]

        for future in concurrent.futures.as_completed(futures):
            try:
                credit_summary = future.result()
                if credit_summary:
                    total_balance += credit_summary.get(
                        "total_balance",
                        0
                    )
            except Exception as exc:
                logging.error(f"获取额度信息生成异常: {exc}")

    return total_balance

@app.route('/handsome/v1/dashboard/billing/usage', methods=['GET'])
def billing_usage():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    daily_usage = []

    return jsonify({
        "object": "list",
        "data": daily_usage,
        "total_usage": 0
    })

@app.route('/handsome/v1/dashboard/billing/subscription', methods=['GET'])
def billing_subscription():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    total_balance = get_billing_info()

    return jsonify({
        "object": "billing_subscription",
        "has_payment_method": False,
        "canceled": False,
        "canceled_at": None,
        "delinquent": None,
        "access_until": int(datetime(9999, 12, 31).timestamp()),
        "soft_limit": 0,
        "hard_limit": total_balance,
        "system_hard_limit": total_balance,
        "soft_limit_usd": 0,
        "hard_limit_usd": total_balance,
        "system_hard_limit_usd": total_balance,
        "plan": {
            "name": "SiliconFlow API",
            "id": "siliconflow-api"
        },
        "account_name": "SiliconFlow User",
        "po_number": None,
        "billing_email": None,
        "tax_ids": [],
        "billing_address": None,
        "business_address": None
    })

@app.route('/handsome/v1/embeddings', methods=['POST'])
def handsome_embeddings():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    data = request.get_json()
    if not data or 'model' not in data:
        return jsonify({"error": "Invalid request data"}), 400

    model_name = data['model']
    request_type = determine_request_type(
        model_name,
        embedding_models,
        free_embedding_models
    )
    api_key = select_key(request_type, model_name)

    if not api_key:
        return jsonify(
            {
                "error": (
                    "No available API key for this "
                    "request type or all keys have "
                    "reached their limits"
                )
            }
        ), 429

    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }

    try:
        start_time = time.time()
        response = requests.post(
            EMBEDDINGS_ENDPOINT,
            headers=headers,
            json=data,
            timeout=120
        )

        if response.status_code == 429:
            return jsonify(response.json()), 429

        response.raise_for_status()
        end_time = time.time()
        response_json = response.json()
        total_time = end_time - start_time

        try:
            prompt_tokens = response_json["usage"]["prompt_tokens"]
            embedding_data = response_json["data"]
        except (KeyError, ValueError, IndexError) as e:
            logging.error(
                f"解析响应 JSON 失败: {e}, "
                f"完整内容: {response_json}"
            )
            prompt_tokens = 0
            embedding_data = []

        logging.info(
            f"使用的key: {api_key}, "
            f"提示token: {prompt_tokens}, "
            f"总共用时: {total_time:.4f}秒, "
            f"使用的模型: {model_name}"
        )

        with data_lock:
            request_timestamps.append(time.time())
            token_counts.append(prompt_tokens)
            request_timestamps_day.append(time.time())
            token_counts_day.append(prompt_tokens)
        
        return jsonify({
            "object": "list",
            "data": embedding_data,
            "model": model_name,
            "usage": {
                "prompt_tokens": prompt_tokens,
                "total_tokens": prompt_tokens
            }
        })

    except requests.exceptions.RequestException as e:
        return jsonify({"error": str(e)}), 500

@app.route('/handsome/v1/images/generations', methods=['POST'])
def handsome_images_generations():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    data = request.get_json()
    if not data or 'model' not in data:
        return jsonify({"error": "Invalid request data"}), 400

    model_name = data.get('model')
    
    request_type = determine_request_type(
        model_name,
        image_models,
        free_image_models
    )
    
    api_key = select_key(request_type, model_name)

    if not api_key:
         return jsonify(
             {
                 "error": (
                     "No available API key for this "
                     "request type or all keys have "
                     "reached their limits"
                 )
             }
         ), 429
    
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    
    response_data = {}
    
    if "stable-diffusion" in model_name or model_name in ["black-forest-labs/FLUX.1-schnell", "Pro/black-forest-labs/FLUX.1-schnell","black-forest-labs/FLUX.1-dev", "black-forest-labs/FLUX.1-pro"]:
        siliconflow_data = {
            "model": model_name,
            "prompt": data.get("prompt"),
            
        }
        
        if model_name == "black-forest-labs/FLUX.1-pro":
            siliconflow_data["width"] = data.get("width", 1024)
            siliconflow_data["height"] = data.get("height", 768)
            siliconflow_data["prompt_upsampling"] = data.get("prompt_upsampling", False)
            siliconflow_data["image_prompt"] = data.get("image_prompt")
            siliconflow_data["steps"] = data.get("steps", 20)
            siliconflow_data["guidance"] = data.get("guidance", 3)
            siliconflow_data["safety_tolerance"] = data.get("safety_tolerance", 2)
            siliconflow_data["interval"] = data.get("interval", 2)
            siliconflow_data["output_format"] = data.get("output_format", "png")
            seed = data.get("seed")
            if isinstance(seed, int) and 0 < seed < 9999999999:
                siliconflow_data["seed"] = seed

            if siliconflow_data["width"] < 256 or siliconflow_data["width"] > 1440 or siliconflow_data["width"] % 32 != 0:
                siliconflow_data["width"] = 1024
            if siliconflow_data["height"] < 256 or siliconflow_data["height"] > 1440 or siliconflow_data["height"] % 32 != 0:
                 siliconflow_data["height"] = 768

            if siliconflow_data["steps"] < 1 or siliconflow_data["steps"] > 50:
                siliconflow_data["steps"] = 20
            if siliconflow_data["guidance"] < 1.5 or siliconflow_data["guidance"] > 5:
                 siliconflow_data["guidance"] = 3
            if siliconflow_data["safety_tolerance"] < 0 or siliconflow_data["safety_tolerance"] > 6:
                siliconflow_data["safety_tolerance"] = 2
            if siliconflow_data["interval"] < 1 or siliconflow_data["interval"] > 4 :
                 siliconflow_data["interval"] = 2
        else:    
            siliconflow_data["image_size"] = data.get("image_size", "1024x1024") 
            siliconflow_data["prompt_enhancement"] = data.get("prompt_enhancement", False)
            seed = data.get("seed")
            if isinstance(seed, int) and 0 < seed < 9999999999:
                siliconflow_data["seed"] = seed
        
            if model_name not in ["black-forest-labs/FLUX.1-schnell", "Pro/black-forest-labs/FLUX.1-schnell"]:
                siliconflow_data["batch_size"] = data.get("n", 1)
                siliconflow_data["num_inference_steps"] = data.get("steps", 20)
                siliconflow_data["guidance_scale"] = data.get("guidance_scale", 7.5)
                siliconflow_data["negative_prompt"] = data.get("negative_prompt")
                if siliconflow_data["batch_size"] < 1:
                    siliconflow_data["batch_size"] = 1
                if siliconflow_data["batch_size"] > 4:
                    siliconflow_data["batch_size"] = 4

                if siliconflow_data["num_inference_steps"] < 1:
                    siliconflow_data["num_inference_steps"] = 1
                if siliconflow_data["num_inference_steps"] > 50:
                    siliconflow_data["num_inference_steps"] = 50
                    
                if siliconflow_data["guidance_scale"] < 0:
                    siliconflow_data["guidance_scale"] = 0
                if siliconflow_data["guidance_scale"] > 100:
                    siliconflow_data["guidance_scale"] = 100

        if "image_size" in siliconflow_data and siliconflow_data["image_size"] not in ["1024x1024", "512x1024", "768x512", "768x1024", "1024x576", "576x1024","960x1280", "720x1440", "720x1280"]:
            siliconflow_data["image_size"] = "1024x1024"

        try:
            start_time = time.time()
            response = requests.post(
                "https://api-st.siliconflow.cn/v1/images/generations",
                headers=headers,
                json=siliconflow_data,
                timeout=120
            )

            if response.status_code == 429:
                return jsonify(response.json()), 429

            response.raise_for_status()
            end_time = time.time()
            response_json = response.json()
            total_time = end_time - start_time
            
            try:
                images = response_json.get("images", [])
                openai_images = []
                for item in images:
                    if isinstance(item, dict) and "url" in item:
                        image_url = item["url"]
                        print(f"image_url: {image_url}")
                        if data.get("response_format") == "b64_json":
                           try:
                                image_data = requests.get(image_url, stream=True).raw
                                image = Image.open(image_data)
                                buffered = io.BytesIO()
                                image.save(buffered, format="PNG")
                                img_str = base64.b64encode(buffered.getvalue()).decode()
                                openai_images.append({"b64_json": img_str})
                           except Exception as e:
                                logging.error(f"图片转base64失败: {e}")
                                openai_images.append({"url": image_url})
                        else:
                            openai_images.append({"url": image_url})
                    else:
                        logging.error(f"无效的图片数据: {item}")
                        openai_images.append({"url": item})


                response_data = {
                    "created": int(time.time()),
                    "data": openai_images
                }
            except (KeyError, ValueError, IndexError) as e:
                logging.error(
                    f"解析响应 JSON 失败: {e}, "
                    f"完整内容: {response_json}"
                )
                response_data = {
                    "created": int(time.time()),
                    "data": []
                }

            logging.info(
                f"使用的key: {api_key}, "
                f"总共用时: {total_time:.4f}秒, "
                f"使用的模型: {model_name}"
            )

            with data_lock:
                request_timestamps.append(time.time())
                token_counts.append(0)
                request_timestamps_day.append(time.time())
                token_counts_day.append(0)

            return jsonify(response_data)

        except requests.exceptions.RequestException as e:
            logging.error(f"请求转发异常: {e}")
            return jsonify({"error": str(e)}), 500
    else:
        return jsonify({"error": "Unsupported model"}), 400

@app.route('/handsome/v1/chat/completions', methods=['POST'])
def handsome_chat_completions():
    if not check_authorization(request):
        return jsonify({"error": "Unauthorized"}), 401

    data = request.get_json()
    if not data or 'model' not in data:
        return jsonify({"error": "Invalid request data"}), 400

    model_name = data['model']
    
    request_type = determine_request_type(
        model_name,
        text_models + image_models,
        free_text_models + free_image_models
    )
    
    api_key = select_key(request_type, model_name)

    if not api_key:
        return jsonify(
            {
                "error": (
                    "No available API key for this "
                    "request type or all keys have "
                    "reached their limits"
                )
            }
        ), 429

    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json"
    }
    
    if model_name in image_models:
        user_content = ""
        messages = data.get("messages", [])
        for message in messages:
            if message["role"] == "user":
                if isinstance(message["content"], str):
                    user_content += message["content"] + " "
                elif isinstance(message["content"], list):
                    for item in message["content"]:
                        if (
                            isinstance(item, dict) and
                            item.get("type") == "text"
                        ):
                            user_content += (
                                item.get("text", "") +
                                " "
                            )
        user_content = user_content.strip()

        siliconflow_data = {
            "model": model_name,
            "prompt": user_content,
            
        }
        if model_name == "black-forest-labs/FLUX.1-pro":
            siliconflow_data["width"] = data.get("width", 1024)
            siliconflow_data["height"] = data.get("height", 768)
            siliconflow_data["prompt_upsampling"] = data.get("prompt_upsampling", False)
            siliconflow_data["image_prompt"] = data.get("image_prompt")
            siliconflow_data["steps"] = data.get("steps", 20)
            siliconflow_data["guidance"] = data.get("guidance", 3)
            siliconflow_data["safety_tolerance"] = data.get("safety_tolerance", 2)
            siliconflow_data["interval"] = data.get("interval", 2)
            siliconflow_data["output_format"] = data.get("output_format", "png")
            seed = data.get("seed")
            if isinstance(seed, int) and 0 < seed < 9999999999:
                siliconflow_data["seed"] = seed
            if siliconflow_data["width"] < 256 or siliconflow_data["width"] > 1440 or siliconflow_data["width"] % 32 != 0:
                siliconflow_data["width"] = 1024
            if siliconflow_data["height"] < 256 or siliconflow_data["height"] > 1440 or siliconflow_data["height"] % 32 != 0:
                 siliconflow_data["height"] = 768

            if siliconflow_data["steps"] < 1 or siliconflow_data["steps"] > 50:
                siliconflow_data["steps"] = 20
            if siliconflow_data["guidance"] < 1.5 or siliconflow_data["guidance"] > 5:
                 siliconflow_data["guidance"] = 3
            if siliconflow_data["safety_tolerance"] < 0 or siliconflow_data["safety_tolerance"] > 6:
                siliconflow_data["safety_tolerance"] = 2
            if siliconflow_data["interval"] < 1 or siliconflow_data["interval"] > 4 :
                 siliconflow_data["interval"] = 2
        else:
            siliconflow_data["image_size"] = "1024x1024"
            siliconflow_data["batch_size"] = 1
            siliconflow_data["num_inference_steps"] = 20
            siliconflow_data["guidance_scale"] = 7.5
            siliconflow_data["prompt_enhancement"] = False
        
            if data.get("size"):
                siliconflow_data["image_size"] = data.get("size")
            if data.get("n"):
                siliconflow_data["batch_size"] = data.get("n")
            if data.get("steps"):
                siliconflow_data["num_inference_steps"] = data.get("steps")
            if data.get("guidance_scale"):
               siliconflow_data["guidance_scale"] = data.get("guidance_scale")
            if data.get("negative_prompt"):
               siliconflow_data["negative_prompt"] = data.get("negative_prompt")
            if data.get("seed"):
               siliconflow_data["seed"] = data.get("seed")
            if data.get("prompt_enhancement"):
               siliconflow_data["prompt_enhancement"] = data.get("prompt_enhancement")

            if siliconflow_data["batch_size"] < 1:
                siliconflow_data["batch_size"] = 1
            if siliconflow_data["batch_size"] > 4:
                siliconflow_data["batch_size"] = 4

            if siliconflow_data["num_inference_steps"] < 1:
                siliconflow_data["num_inference_steps"] = 1
            if siliconflow_data["num_inference_steps"] > 50:
                siliconflow_data["num_inference_steps"] = 50
                
            if siliconflow_data["guidance_scale"] < 0:
                siliconflow_data["guidance_scale"] = 0
            if siliconflow_data["guidance_scale"] > 100:
                siliconflow_data["guidance_scale"] = 100
            
            if siliconflow_data["image_size"] not in ["1024x1024", "512x1024", "768x512", "768x1024", "1024x576", "576x1024", "960x1280", "720x1440", "720x1280"]:
                siliconflow_data["image_size"] = "1024x1024"

        try:
            start_time = time.time()
            response = requests.post(
                "https://api.siliconflow.cn/v1/images/generations",
                headers=headers,
                json=siliconflow_data,
                timeout=120,
                stream=data.get("stream", False)
            )
           
            if response.status_code == 429:
                return jsonify(response.json()), 429

            if data.get("stream", False):
                def generate():
                    try:
                        response.raise_for_status()
                        response_json = response.json()
                    
                        images = response_json.get("images", [])
                        
                        image_url = ""
                        if images and isinstance(images[0], dict) and "url" in images[0]:
                            image_url = images[0]["url"]
                            logging.info(f"Extracted image URL: {image_url}")
                        elif images and isinstance(images[0], str):
                            image_url = images[0]
                            logging.info(f"Extracted image URL: {image_url}")
                        
                        markdown_image_link = f"![image]({image_url})"
                        if image_url:
                            chunk_data = {
                                "id": f"chatcmpl-{uuid.uuid4()}",
                                "object": "chat.completion.chunk",
                                "created": int(time.time()),
                                "model": model_name,
                                "choices": [
                                    {
                                        "index": 0,
                                        "delta": {
                                            "role": "assistant",
                                            "content": markdown_image_link
                                        },
                                        "finish_reason": None
                                    }
                                ]
                            }
                            yield f"data: {json.dumps(chunk_data)}\n\n".encode('utf-8')
                        else:
                            chunk_data = {
                                "id": f"chatcmpl-{uuid.uuid4()}",
                                "object": "chat.completion.chunk",
                                "created": int(time.time()),
                                "model": model_name,
                                "choices": [
                                    {
                                        "index": 0,
                                        "delta": {
                                            "role": "assistant",
                                            "content": "Failed to generate image"
                                        },
                                        "finish_reason": None
                                    }
                                ]
                            }
                            yield f"data: {json.dumps(chunk_data)}\n\n".encode('utf-8')
                        
                        end_chunk_data = {
                            "id": f"chatcmpl-{uuid.uuid4()}",
                            "object": "chat.completion.chunk",
                            "created": int(time.time()),
                            "model": model_name,
                            "choices": [
                                {
                                    "index": 0,
                                    "delta": {},
                                    "finish_reason": "stop"
                                }
                            ]
                        }
                        yield f"data: {json.dumps(end_chunk_data)}\n\n".encode('utf-8')
                        with data_lock:
                            request_timestamps.append(time.time())
                            token_counts.append(0)
                            request_timestamps_day.append(time.time())
                            token_counts_day.append(0)
                    except requests.exceptions.RequestException as e:
                        logging.error(f"请求转发异常: {e}")
                        error_chunk_data = {
                            "id": f"chatcmpl-{uuid.uuid4()}",
                            "object": "chat.completion.chunk",
                            "created": int(time.time()),
                            "model": model_name,
                            "choices": [
                                {
                                    "index": 0,
                                    "delta": {
                                        "role": "assistant",
                                        "content": f"Error: {str(e)}"
                                    },
                                    "finish_reason": None
                                }
                            ]
                        }
                        yield f"data: {json.dumps(error_chunk_data)}\n\n".encode('utf-8')
                        end_chunk_data = {
                                "id": f"chatcmpl-{uuid.uuid4()}",
                                "object": "chat.completion.chunk",
                                "created": int(time.time()),
                                "model": model_name,
                                "choices": [
                                    {
                                        "index": 0,
                                        "delta": {},
                                        "finish_reason": "stop"
                                    }
                                ]
                            }
                        yield f"data: {json.dumps(end_chunk_data)}\n\n".encode('utf-8')
                    logging.info(
                        f"使用的key: {api_key}, "
                        f"使用的模型: {model_name}"
                    )
                    yield "data: [DONE]\n\n".encode('utf-8')
                return Response(stream_with_context(generate()), content_type='text/event-stream')

            else:
                response.raise_for_status()
                end_time = time.time()
                response_json = response.json()
                total_time = end_time - start_time
            
                try:
                    images = response_json.get("images", [])
                    
                    image_url = ""
                    if images and isinstance(images[0], dict) and "url" in images[0]:
                        image_url = images[0]["url"]
                        logging.info(f"Extracted image URL: {image_url}")
                    elif images and isinstance(images[0], str):
                        image_url = images[0]
                        logging.info(f"Extracted image URL: {image_url}")
                    
                    markdown_image_link = f"![image]({image_url})"
                    response_data = {
                        "id": f"chatcmpl-{uuid.uuid4()}",
                        "object": "chat.completion",
                        "created": int(time.time()),
                        "model": model_name,
                        "choices": [
                            {
                            "index": 0,
                            "message": {
                                "role": "assistant",
                                "content": markdown_image_link if image_url else "Failed to generate image",
                            },
                            "finish_reason": "stop",
                            }
                        ],
                    }
                except (KeyError, ValueError, IndexError) as e:
                    logging.error(
                        f"解析响应 JSON 失败: {e}, "
                        f"完整内容: {response_json}"
                    )
                    response_data = {
                        "id": f"chatcmpl-{uuid.uuid4()}",
                        "object": "chat.completion",
                        "created": int(time.time()),
                        "model": model_name,
                        "choices": [
                            {
                            "index": 0,
                            "message": {
                                "role": "assistant",
                                "content": "Failed to process image data",
                            },
                            "finish_reason": "stop",
                            }
                        ],
                    }

                logging.info(
                    f"使用的key: {api_key}, "
                    f"总共用时: {total_time:.4f}秒, "
                    f"使用的模型: {model_name}"
                )
                with data_lock:
                    request_timestamps.append(time.time())
                    token_counts.append(0)
                    request_timestamps_day.append(time.time())
                    token_counts_day.append(0)
                return jsonify(response_data)

        except requests.exceptions.RequestException as e:
            logging.error(f"请求转发异常: {e}")
            return jsonify({"error": str(e)}), 500
    else:
        try:
            start_time = time.time()
            response = requests.post(
                TEST_MODEL_ENDPOINT,
                headers=headers,
                json=data,
                stream=data.get("stream", False),
                timeout=60
            )

            if response.status_code == 429:
                return jsonify(response.json()), 429

            if data.get("stream", False):
                def generate():
                    first_chunk_time = None
                    full_response_content = ""
                    for chunk in response.iter_content(chunk_size=2048):
                        if chunk:
                            if first_chunk_time is None:
                                first_chunk_time = time.time()
                            full_response_content += chunk.decode("utf-8")
                            yield chunk

                    end_time = time.time()
                    first_token_time = (
                        first_chunk_time - start_time
                        if first_chunk_time else 0
                    )
                    total_time = end_time - start_time

                    prompt_tokens = 0
                    completion_tokens = 0
                    response_content = ""
                    for line in full_response_content.splitlines():
                        if line.startswith("data:"):
                            line = line[5:].strip()
                            if line == "[DONE]":
                                continue
                            try:
                                response_json = json.loads(line)

                                if (
                                    "usage" in response_json and
                                    "completion_tokens" in response_json["usage"]
                                ):
                                    completion_tokens = response_json[
                                        "usage"
                                    ]["completion_tokens"]

                                if (
                                    "choices" in response_json and
                                    len(response_json["choices"]) > 0 and
                                    "delta" in response_json["choices"][0] and
                                    "content" in response_json[
                                        "choices"
                                    ][0]["delta"]
                                ):
                                    response_content += response_json[
                                        "choices"
                                    ][0]["delta"]["content"]

                                if (
                                    "usage" in response_json and
                                    "prompt_tokens" in response_json["usage"]
                                ):
                                    prompt_tokens = response_json[
                                        "usage"
                                    ]["prompt_tokens"]

                            except (
                                KeyError,
                                ValueError,
                                IndexError
                            ) as e:
                                logging.error(
                                    f"解析流式响应单行 JSON 失败: {e}, "
                                    f"行内容: {line}"
                                )

                    user_content = ""
                    messages = data.get("messages", [])
                    for message in messages:
                        if message["role"] == "user":
                            if isinstance(message["content"], str):
                                user_content += message["content"] + " "
                            elif isinstance(message["content"], list):
                                for item in message["content"]:
                                    if (
                                        isinstance(item, dict) and
                                        item.get("type") == "text"
                                    ):
                                        user_content += (
                                            item.get("text", "") +
                                            " "
                                        )

                    user_content = user_content.strip()

                    user_content_replaced = user_content.replace(
                        '\n', '\\n'
                    ).replace('\r', '\\n')
                    response_content_replaced = response_content.replace(
                        '\n', '\\n'
                    ).replace('\r', '\\n')

                    logging.info(
                        f"使用的key: {api_key}, "
                        f"提示token: {prompt_tokens}, "
                        f"输出token: {completion_tokens}, "
                        f"首字用时: {first_token_time:.4f}秒, "
                        f"总共用时: {total_time:.4f}秒, "
                        f"使用的模型: {model_name}, "
                        f"用户的内容: {user_content_replaced}, "
                        f"输出的内容: {response_content_replaced}"
                    )

                    with data_lock:
                        request_timestamps.append(time.time())
                        token_counts.append(prompt_tokens+completion_tokens)
                        request_timestamps_day.append(time.time())
                        token_counts_day.append(prompt_tokens+completion_tokens)

                return Response(
                    stream_with_context(generate()),
                    content_type=response.headers['Content-Type']
                )
            else:
                response.raise_for_status()
                end_time = time.time()
                response_json = response.json()
                total_time = end_time - start_time

                try:
                    prompt_tokens = response_json["usage"]["prompt_tokens"]
                    completion_tokens = response_json[
                        "usage"
                    ]["completion_tokens"]
                    response_content = response_json[
                        "choices"
                    ][0]["message"]["content"]
                except (KeyError, ValueError, IndexError) as e:
                    logging.error(
                        f"解析非流式响应 JSON 失败: {e}, "
                        f"完整内容: {response_json}"
                    )
                    prompt_tokens = 0
                    completion_tokens = 0
                    response_content = ""

                user_content = ""
                messages = data.get("messages", [])
                for message in messages:
                    if message["role"] == "user":
                        if isinstance(message["content"], str):
                            user_content += message["content"] + " "
                        elif isinstance(message["content"], list):
                            for item in message["content"]:
                                if (
                                    isinstance(item, dict) and
                                    item.get("type") == "text"
                                ):
                                    user_content += (
                                        item.get("text", "") + " "
                                    )

                user_content = user_content.strip()

                user_content_replaced = user_content.replace(
                    '\n', '\\n'
                ).replace('\r', '\\n')
                response_content_replaced = response_content.replace(
                    '\n', '\\n'
                ).replace('\r', '\\n')

                logging.info(
                    f"使用的key: {api_key}, "
                    f"提示token: {prompt_tokens}, "
                    f"输出token: {completion_tokens}, "
                    f"首字用时: 0, "
                    f"总共用时: {total_time:.4f}秒, "
                    f"使用的模型: {model_name}, "
                    f"用户的内容: {user_content_replaced}, "
                    f"输出的内容: {response_content_replaced}"
                )
                with data_lock:
                    request_timestamps.append(time.time())
                    if "prompt_tokens" in response_json["usage"] and "completion_tokens" in response_json["usage"]:
                        token_counts.append(response_json["usage"]["prompt_tokens"] + response_json["usage"]["completion_tokens"])
                    else:
                        token_counts.append(0)
                    request_timestamps_day.append(time.time())
                    if "prompt_tokens" in response_json["usage"] and "completion_tokens" in response_json["usage"]:
                        token_counts_day.append(response_json["usage"]["prompt_tokens"] + response_json["usage"]["completion_tokens"])
                    else:
                        token_counts_day.append(0)

                return jsonify(response_json)

        except requests.exceptions.RequestException as e:
            logging.error(f"请求转发异常: {e}")
            return jsonify({"error": str(e)}), 500

if __name__ == '__main__':
    logging.info(f"环境变量：{os.environ}")

    load_keys()
    logging.info("程序启动时首次加载 keys 已执行")

    scheduler.start()

    logging.info("首次加载 keys 已手动触发执行")

    refresh_models()
    logging.info("首次刷新模型列表已手动触发执行")

    app.run(
        debug=False,
        host='0.0.0.0',
        port=int(os.environ.get('PORT', 7860))
    )