from flask import Flask, request, jsonify, Response, stream_with_context, render_template_string
from google.generativeai.types import generation_types
from google.api_core.exceptions import InvalidArgument, ResourceExhausted, ServiceUnavailable, InternalServerError, Aborted
import google.generativeai as genai
import json
import os
import logging
import func
from datetime import datetime, timedelta
from apscheduler.schedulers.background import BackgroundScheduler
import time
import requests
from collections import deque
os.environ['TZ'] = 'Asia/Shanghai'
app = Flask(__name__)
app.secret_key = os.urandom(24)
formatter = logging.Formatter('%(message)s')
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
handler = logging.StreamHandler()
handler.setFormatter(formatter)
logger.addHandler(handler)
MAX_RETRIES = int(os.environ.get('MaxRetries', 3))
MAX_REQUESTS = int(os.environ.get('MaxRequests', 2))
LIMIT_WINDOW = int(os.environ.get('LimitWindow', 60))
RETRY_DELAY = 1
MAX_RETRY_DELAY = 16
request_counts = {}
api_key_blacklist = set()
api_key_blacklist_duration = 60
# 核心优势
safety_settings = [
{
"category": "HARM_CATEGORY_HARASSMENT",
"threshold": "BLOCK_NONE"
},
{
"category": "HARM_CATEGORY_HATE_SPEECH",
"threshold": "BLOCK_NONE"
},
{
"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
"threshold": "BLOCK_NONE"
},
{
"category": "HARM_CATEGORY_DANGEROUS_CONTENT",
"threshold": "BLOCK_NONE"
},
]
class APIKeyManager:
def __init__(self):
self.api_keys = os.environ.get('KeyArray').split(',')
self.current_index = 0
def get_available_key(self):
num_keys = len(self.api_keys)
for _ in range(num_keys):
if self.current_index >= num_keys:
self.current_index = 0
current_key = self.api_keys[self.current_index]
self.current_index += 1
if current_key not in api_key_blacklist:
return current_key
logger.error("所有API key都已耗尽或被暂时禁用，请重新配置或稍后重试")
return None
def show_all_keys(self):
logger.info(f"当前可用API key个数: {len(self.api_keys)} ")
for i, api_key in enumerate(self.api_keys):
logger.info(f"API Key{i}: {api_key[:11]}...")
def blacklist_key(self, key):
logger.warning(f"{key[:11]} → 暂时禁用 {api_key_blacklist_duration} 秒")
api_key_blacklist.add(key)
scheduler.add_job(lambda: api_key_blacklist.discard(key), 'date', run_date=datetime.now() + timedelta(seconds=api_key_blacklist_duration))
key_manager = APIKeyManager()
key_manager.show_all_keys()
current_api_key = key_manager.get_available_key()
def switch_api_key():
global current_api_key
key = key_manager.get_available_key()
if key:
current_api_key = key
logger.info(f"API key 替换为 → {current_api_key[:11]}...")
else:
logger.error("API key 替换失败，所有API key都已耗尽或被暂时禁用，请重新配置或稍后重试")
logger.info(f"当前 API key: {current_api_key[:11]}...")
GEMINI_MODELS = [
{"id": "gemini-1.5-pro-002"},
{"id": "gemini-1.5-pro-latest"},
{"id": "gemini-1.5-pro-exp-0827"},
{"id": "learnlm-1.5-pro-experimental"},
{"id": "gemini-exp-1114"},
{"id": "gemini-exp-1121"},
{"id": "gemini-exp-1206"},
{"id": "gemini-2.0-flash-exp"},
{"id": "gemini-2.0-flash-thinking-exp-1219"},
{"id": "gemini-2.0-pro-exp"}
]
@app.route('/')
def index():
main_content = "Moonfanz Reminiproxy v2.1.2 2025-01-09"
html_template = """
{{ main_content }}
完全开源、免费且禁止商用
点击复制反向代理: Copy Link
聊天来源选择"自定义(兼容 OpenAI)"
将复制的网址填入到自定义端点
将设置password填入自定义API秘钥
"""
return render_template_string(html_template, main_content=main_content)
def is_within_rate_limit(api_key):
now = datetime.now()
if api_key not in request_counts:
request_counts[api_key] = deque()
while request_counts[api_key] and request_counts[api_key][0] < now - timedelta(seconds=LIMIT_WINDOW):
request_counts[api_key].popleft()
if len(request_counts[api_key]) >= MAX_REQUESTS:
earliest_request_time = request_counts[api_key][0]
wait_time = (earliest_request_time + timedelta(seconds=LIMIT_WINDOW)) - now
return False, wait_time.total_seconds()
else:
return True, 0
def increment_request_count(api_key):
now = datetime.now()
if api_key not in request_counts:
request_counts[api_key] = deque()
request_counts[api_key].append(now)
def handle_api_error(error, attempt, stream=False):
if attempt > MAX_RETRIES:
logger.error(f"{MAX_RETRIES} 次尝试后仍然失败，请修改预设或输入")
return False, jsonify({
'error': {
'message': f"{MAX_RETRIES} 次尝试后仍然失败，请修改预设或输入",
'type': 'max_retries_exceeded'
}
})
if isinstance(error, InvalidArgument):
logger.error(f"{current_api_key[:11]} → 无效，可能已过期或被删除")
key_manager.blacklist_key(current_api_key)
switch_api_key()
return False, None
elif isinstance(error, (ResourceExhausted, Aborted, InternalServerError, ServiceUnavailable)):
delay = min(RETRY_DELAY * (2 ** attempt), MAX_RETRY_DELAY)
if isinstance(error, ResourceExhausted):
logger.warning(f"{current_api_key[:11]} → 429 官方资源耗尽 → {delay} 秒后重试...")
else:
logger.warning(f"{current_api_key[:11]} → 未知错误↙ {delay} 秒后重试...\n{type(error).__name__}\n")
key_manager.blacklist_key(current_api_key)
switch_api_key()
time.sleep(delay)
return False, None
elif isinstance(error, generation_types.StopCandidateException):
logger.warning(f"AI输出内容被Gemini官方阻挡，代理没有得到有效回复")
switch_api_key()
return False, None
else:
logger.error(f"未知错误↙\n {error}")
return False, None
@app.route('/hf/v1/chat/completions', methods=['POST'])
def chat_completions():
is_authenticated, auth_error, status_code = func.authenticate_request(request)
if not is_authenticated:
return auth_error if auth_error else jsonify({'error': '未授权'}), status_code if status_code else 401
request_data = request.get_json()
messages = request_data.get('messages', [])
model = request_data.get('model', 'gemini-2.0-flash-exp')
temperature = request_data.get('temperature', 1)
max_tokens = request_data.get('max_tokens', 8192)
stream = request_data.get('stream', False)
logger.info(f"\n{model} [r] → {current_api_key[:11]}...")
gemini_history, user_message, error_response = func.process_messages_for_gemini(messages)
if error_response:
logger.error(f"处理输入消息时出错↙\n {error_response}")
return jsonify(error_response), 400
def do_request(current_api_key, attempt):
isok, time = is_within_rate_limit(current_api_key)
if not isok:
logger.warning(f"{current_api_key[:11]} → 暂时超过限额，该API key将在 {time} 秒后启用...")
switch_api_key()
return False, None
increment_request_count(current_api_key)
genai.configure(api_key=current_api_key)
generation_config = {
"temperature": temperature,
"max_output_tokens": max_tokens
}
gen_model = genai.GenerativeModel(
model_name=model,
generation_config=generation_config,
safety_settings=safety_settings
)
try:
if gemini_history:
chat_session = gen_model.start_chat(history=gemini_history)
response = chat_session.send_message(user_message, stream=stream)
else:
response = gen_model.generate_content(user_message, stream=stream)
return True, response
except Exception as e:
return handle_api_error(e, attempt, stream)
def generate(response):
try:
for chunk in response:
if chunk.text:
data = {
'choices': [
{
'delta': {
'content': chunk.text
},
'finish_reason': None,
'index': 0
}
],
'object': 'chat.completion.chunk'
}
yield f"data: {json.dumps(data)}\n\n"
data = {
'choices': [
{
'delta': {},
'finish_reason': 'stop',
'index': 0
}
],
'object': 'chat.completion.chunk'
}
yield f"data: {json.dumps(data)}\n\n"
logger.info(f"200!")
except Exception:
logger.error(f"流式输出时截断，请关闭流式输出或修改你的输入")
error_data = {
'error': {
'message': '流式输出时截断，请关闭流式输出或修改你的输入',
'type': 'internal_server_error'
}
}
yield f"data: {json.dumps(error_data)}\n\n"
data = {
'choices': [
{
'delta': {},
'finish_reason': 'stop',
'index': 0
}
],
'object': 'chat.completion.chunk'
}
yield f"data: {json.dumps(data)}\n\n"
attempt = 0
success = False
response = None
while attempt < MAX_RETRIES and not success:
attempt += 1
logger.info(f"第 {attempt}/{MAX_RETRIES} 次尝试 ...")
success, response = do_request(current_api_key, attempt)
if not success:
logger.error(f"{MAX_RETRIES} 次尝试均失败，请调整配置或向Moonfanz反馈")
response = {
'error': {
'message': f'{MAX_RETRIES} 次尝试均失败，请调整配置或向Moonfanz反馈',
'type': 'internal_server_error'
}
}
return jsonify(response), 500 if response is not None else 503
if stream:
return Response(stream_with_context(generate(response)), mimetype='text/event-stream')
else:
try:
text_content = response.text
except (AttributeError, IndexError, TypeError, ValueError) as e:
if "response.candidates is empty" in str(e):
logger.error(f"你的输入被AI安全过滤器阻止")
return jsonify({
'error': {
'message': '你的输入被AI安全过滤器阻止',
'type': 'prompt_blocked_error',
'details': str(e)
}
}), 400
else:
logger.error(f"AI响应处理失败")
return jsonify({
'error': {
'message': 'AI响应处理失败',
'type': 'response_processing_error'
}
}), 500
response_data = {
'id': 'chatcmpl-xxxxxxxxxxxx',
'object': 'chat.completion',
'created': int(datetime.now().timestamp()),
'model': model,
'choices': [{
'index': 0,
'message': {
'role': 'assistant',
'content': text_content
},
'finish_reason': 'stop'
}],
'usage': {
'prompt_tokens': 0,
'completion_tokens': 0,
'total_tokens': 0
}
}
logger.info(f"200!")
return jsonify(response_data)
@app.route('/hf/v1/models', methods=['GET'])
def list_models():
response = {"object": "list", "data": GEMINI_MODELS}
return jsonify(response)
def keep_alive():
try:
response = requests.get("http://127.0.0.1:7860/", timeout=10)
response.raise_for_status()
print(f"Keep alive ping successful: {response.status_code} at {time.ctime()}")
except requests.exceptions.RequestException as e:
print(f"Keep alive ping failed: {e} at {time.ctime()}")
if __name__ == '__main__':
scheduler = BackgroundScheduler()
scheduler.add_job(keep_alive, 'interval', hours=12)
scheduler.start()
logger.info(f"最大尝试次数/MaxRetries: {MAX_RETRIES}")
logger.info(f"最大请求次数/MaxRequests: {MAX_REQUESTS}")
logger.info(f"请求限额窗口/LimitWindow: {LIMIT_WINDOW} 秒")
app.run(debug=True, host='0.0.0.0', port=int(os.environ.get('PORT', 7860)))