|
import os |
|
import time |
|
import logging |
|
import requests |
|
import json |
|
import concurrent.futures |
|
import threading |
|
from datetime import datetime, timedelta |
|
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.deepseek.com/user/balance" |
|
TEST_MODEL_ENDPOINT = "https://api.deepseek.com/v1/chat/completions" |
|
MODELS_ENDPOINT = "https://api.deepseek.com/models" |
|
|
|
app = Flask(__name__) |
|
|
|
text_models = [] |
|
|
|
invalid_keys_global = [] |
|
valid_keys_global = [] |
|
|
|
executor = concurrent.futures.ThreadPoolExecutor(max_workers=10000) |
|
model_key_indices = {} |
|
|
|
request_timestamps = [] |
|
token_counts = [] |
|
data_lock = threading.Lock() |
|
|
|
def get_credit_summary(api_key): |
|
headers = { |
|
"Authorization": f"Bearer {api_key}", |
|
"Content-Type": "application/json" |
|
} |
|
try: |
|
response = requests.get(API_ENDPOINT, headers=headers) |
|
response.raise_for_status() |
|
data = response.json() |
|
if not data.get("is_available", False): |
|
logging.warning(f"API Key: {api_key} is not available.") |
|
return None |
|
|
|
balance_infos = data.get("balance_infos", []) |
|
total_balance_cny = 0.0 |
|
usd_balance = 0.0 |
|
for balance_info in balance_infos: |
|
currency = balance_info.get("currency") |
|
total_balance = float(balance_info.get("total_balance", 0)) |
|
|
|
if currency == "CNY": |
|
total_balance_cny += total_balance |
|
elif currency == "USD": |
|
usd_balance = total_balance |
|
|
|
try: |
|
exchange_rate = get_usd_to_cny_rate() |
|
if exchange_rate is not None: |
|
total_balance_cny += usd_balance * exchange_rate |
|
logging.info(f"获取美元兑人民币汇率成功，API Key：{api_key}，当前总额度(CNY): {total_balance_cny}") |
|
else: |
|
logging.warning(f"获取美元兑人民币汇率失败，无法转换美元余额，API Key：{api_key}") |
|
total_balance_cny += usd_balance * 7.2 |
|
except Exception as e: |
|
logging.error(f"获取美元兑人民币汇率失败，API Key：{api_key}，错误信息：{e}") |
|
total_balance_cny += usd_balance * 7.2 |
|
|
|
return {"total_balance": float(total_balance_cny)} |
|
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 |
|
except Exception as e: |
|
logging.error(f"处理额度信息失败，API Key：{api_key}，错误信息：{e}") |
|
return None |
|
|
|
def get_usd_to_cny_rate(): |
|
try: |
|
response = requests.get("https://api.exchangerate-api.com/v4/latest/USD") |
|
response.raise_for_status() |
|
data = response.json() |
|
return data.get("rates", {}).get("CNY") |
|
except requests.exceptions.RequestException as e: |
|
logging.error(f"获取美元兑人民币汇率失败，错误信息：{e}") |
|
return None |
|
|
|
def refresh_models(): |
|
text_models = ["deepseek-chat", "deepseek-reasoner"] |
|
logging.info(f"所有文本模型列表：{text_models}") |
|
|
|
def load_keys(): |
|
keys_str = os.environ.get("KEYS") |
|
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 |
|
): key for key in unique_keys |
|
} |
|
|
|
invalid_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 == "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"有效 KEY：{valid_keys}") |
|
|
|
global invalid_keys_global, valid_keys_global |
|
invalid_keys_global = invalid_keys |
|
valid_keys_global = valid_keys |
|
|
|
def process_key(key): |
|
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: |
|
return "invalid" |
|
else: |
|
return "valid" |
|
|
|
def select_key(model_name): |
|
available_keys = valid_keys_global |
|
|
|
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 |
|
model_key_indices[model_name] = current_index |
|
return key |
|
|
|
model_key_indices[model_name] = 0 |
|
return None |
|
|
|
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() |
|
|
|
@app.route('/') |
|
def index(): |
|
current_time = time.time() |
|
one_minute_ago = current_time - 60 |
|
|
|
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) |
|
|
|
return jsonify({"rpm": rpm, "tpm": tpm}) |
|
|
|
@app.route('/handsome/v1/models', methods=['GET']) |
|
def list_models(): |
|
if not check_authorization(request): |
|
return jsonify({"error": "Unauthorized"}), 401 |
|
|
|
detailed_models = [ |
|
{ |
|
"id": "deepseek-chat", |
|
"object": "model", |
|
"created": 1678888888, |
|
"owned_by": "openai", |
|
"root": "deepseek-chat", |
|
"parent": None |
|
}, |
|
{ |
|
"id": "deepseek-reasoner", |
|
"object": "model", |
|
"created": 1678888889, |
|
"owned_by": "openai", |
|
"root": "deepseek-reasoner", |
|
"parent": None |
|
}, |
|
{ |
|
"id": "deepseek-reasoner-thinking", |
|
"object": "model", |
|
"created": 1678888889, |
|
"owned_by": "openai", |
|
"root": "deepseek-reasoner", |
|
"parent": None |
|
}, |
|
{ |
|
"id": "deepseek-reasoner-openwebui", |
|
"object": "model", |
|
"created": 1678888889, |
|
"owned_by": "openai", |
|
"root": "deepseek-reasoner", |
|
"parent": None |
|
} |
|
] |
|
|
|
return jsonify({ |
|
"success": True, |
|
"data": detailed_models |
|
}) |
|
|
|
def get_billing_info(): |
|
keys = valid_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 |
|
|
|
end_date = datetime.now() |
|
start_date = end_date - timedelta(days=30) |
|
|
|
daily_usage = [] |
|
current_date = start_date |
|
while current_date <= end_date: |
|
daily_usage.append({ |
|
"timestamp": int(current_date.timestamp()), |
|
"daily_usage": 0 |
|
}) |
|
current_date += timedelta(days=1) |
|
|
|
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/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'] |
|
if model_name == "deepseek-reasoner-thinking" or model_name == "deepseek-reasoner-openwebui": |
|
model_realname = "deepseek-reasoner" |
|
else: |
|
model_realname = model_name |
|
|
|
data['model'] = model_realname |
|
|
|
api_key = select_key(model_realname) |
|
|
|
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( |
|
TEST_MODEL_ENDPOINT, |
|
headers=headers, |
|
json=data, |
|
stream=data.get("stream", False), |
|
timeout=120 |
|
) |
|
|
|
if response.status_code == 429: |
|
return jsonify(response.json()), 429 |
|
|
|
if data.get("stream", False): |
|
def generate(): |
|
if model_name == "deepseek-reasoner": |
|
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 = extract_user_content(data.get("messages", [])) |
|
|
|
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'] |
|
) |
|
|
|
if model_name == "deepseek-reasoner-openwebui": |
|
first_chunk_time = None |
|
full_response_content = "" |
|
reasoning_content_accumulated = "" |
|
content_accumulated = "" |
|
first_reasoning_chunk = True |
|
|
|
for chunk in response.iter_lines(): |
|
if chunk: |
|
if first_chunk_time is None: |
|
first_chunk_time = time.time() |
|
full_response_content += chunk.decode("utf-8") |
|
|
|
for line in chunk.decode("utf-8").splitlines(): |
|
if line.startswith("data:"): |
|
try: |
|
chunk_json = json.loads(line.lstrip("data: ").strip()) |
|
if "choices" in chunk_json and len(chunk_json["choices"]) > 0: |
|
delta = chunk_json["choices"][0].get("delta", {}) |
|
|
|
if delta.get("reasoning_content") is not None: |
|
reasoning_chunk = delta["reasoning_content"] |
|
if first_reasoning_chunk: |
|
think_chunk = f"<" |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': think_chunk}, 'index': 0}]})}\n\n" |
|
think_chunk = f"think" |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': think_chunk}, 'index': 0}]})}\n\n" |
|
think_chunk = f">\n" |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': think_chunk}, 'index': 0}]})}\n\n" |
|
first_reasoning_chunk = False |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': reasoning_chunk}, 'index': 0}]})}\n\n" |
|
|
|
if delta.get("content") is not None: |
|
if not first_reasoning_chunk: |
|
reasoning_chunk = f"\n</think>\n" |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': reasoning_chunk}, 'index': 0}]})}\n\n" |
|
first_reasoning_chunk = True |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': delta["content"]}, 'index': 0}]})}\n\n" |
|
|
|
except (KeyError, ValueError, json.JSONDecodeError) as e: |
|
continue |
|
|
|
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 |
|
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 ( |
|
"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 = (f"```Thinking\n{reasoning_content_accumulated}\n```\n" if reasoning_content_accumulated else "") + content_accumulated |
|
response_content_replaced = response_content_replaced.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) |
|
|
|
yield "data: [DONE]\n\n" |
|
|
|
return Response( |
|
stream_with_context(generate()), |
|
content_type="text/event-stream" |
|
) |
|
|
|
first_chunk_time = None |
|
full_response_content = "" |
|
reasoning_content_accumulated = "" |
|
content_accumulated = "" |
|
first_reasoning_chunk = True |
|
|
|
for chunk in response.iter_lines(): |
|
if chunk: |
|
if first_chunk_time is None: |
|
first_chunk_time = time.time() |
|
full_response_content += chunk.decode("utf-8") |
|
|
|
for line in chunk.decode("utf-8").splitlines(): |
|
if line.startswith("data:"): |
|
try: |
|
chunk_json = json.loads(line.lstrip("data: ").strip()) |
|
if "choices" in chunk_json and len(chunk_json["choices"]) > 0: |
|
delta = chunk_json["choices"][0].get("delta", {}) |
|
|
|
if delta.get("reasoning_content") is not None: |
|
reasoning_chunk = delta["reasoning_content"] |
|
reasoning_chunk = reasoning_chunk.replace('\n', '\n> ') |
|
if first_reasoning_chunk: |
|
reasoning_chunk = "> " + reasoning_chunk |
|
first_reasoning_chunk = False |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': reasoning_chunk}, 'index': 0}]})}\n\n" |
|
|
|
if delta.get("content") is not None: |
|
if not first_reasoning_chunk: |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': '\n\n'}, 'index': 0}]})}\n\n" |
|
first_reasoning_chunk = True |
|
yield f"data: {json.dumps({'choices': [{'delta': {'content': delta["content"]}, 'index': 0}]})}\n\n" |
|
|
|
except (KeyError, ValueError, json.JSONDecodeError) as e: |
|
|
|
continue |
|
|
|
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 |
|
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 ( |
|
"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 = (f"```Thinking\n{reasoning_content_accumulated}\n```\n" if reasoning_content_accumulated else "") + content_accumulated |
|
response_content_replaced = response_content_replaced.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) |
|
|
|
yield "data: [DONE]\n\n" |
|
|
|
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: |
|
prompt_tokens = response_json["usage"]["prompt_tokens"] |
|
completion_tokens = response_json["usage"]["completion_tokens"] |
|
response_content = "" |
|
|
|
if model_name == "deepseek-reasoner-thinking" and "choices" in response_json and len(response_json["choices"]) > 0: |
|
choice = response_json["choices"][0] |
|
if "message" in choice: |
|
if "reasoning_content" in choice["message"]: |
|
reasoning_content = choice["message"]["reasoning_content"] |
|
reasoning_content = reasoning_content.replace('\n', '\n> ') |
|
reasoning_content = '> ' + reasoning_content |
|
formatted_reasoning = f"{reasoning_content}\n" |
|
response_content += formatted_reasoning + "\n" |
|
if "content" in choice["message"]: |
|
response_content += choice["message"]["content"] |
|
elif model_name == "deepseek-reasoner-openwebui" and "choices" in response_json and len(response_json["choices"]) > 0: |
|
choice = response_json["choices"][0] |
|
if "message" in choice: |
|
if "reasoning_content" in choice["message"]: |
|
reasoning_content = choice["message"]["reasoning_content"] |
|
response_content += f"<think>\n{reasoning_content}\n</think>\n" |
|
if "content" in choice["message"]: |
|
response_content += choice["message"]["content"] |
|
elif "choices" in response_json and len(response_json["choices"]) > 0: |
|
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()) |
|
token_counts.append(prompt_tokens + completion_tokens) |
|
|
|
formatted_response = { |
|
"id": response_json.get("id", ""), |
|
"object": "chat.completion", |
|
"created": response_json.get("created", int(time.time())), |
|
"model": model_name, |
|
"choices": [ |
|
{ |
|
"index": 0, |
|
"message": { |
|
"role": "assistant", |
|
"content": response_content |
|
}, |
|
"finish_reason": "stop" |
|
} |
|
], |
|
"usage": { |
|
"prompt_tokens": prompt_tokens, |
|
"completion_tokens": completion_tokens, |
|
"total_tokens": prompt_tokens + completion_tokens |
|
} |
|
} |
|
|
|
if model_name == "deepseek-reasoner": |
|
formatted_response = response_json |
|
|
|
return jsonify(formatted_response) |
|
|
|
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}") |
|
|
|
invalid_keys_global = [] |
|
valid_keys_global = [] |
|
|
|
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)) |
|
) |
