mirror of
https://github.com/snailyp/gemini-balance.git
synced 2026-06-03 14:49:42 +08:00
初始提交
This commit is contained in:
yinpeng
246
main.py
246
main.py
@@ -40,40 +40,56 @@ API_KEYS = config.settings.API_KEYS
|
||||
|
||||
# 创建一个循环迭代器
|
||||
key_cycle = cycle(API_KEYS)
|
||||
key_lock = asyncio.Lock()
|
||||
|
||||
# 创建两个独立的锁
|
||||
key_cycle_lock = asyncio.Lock()
|
||||
failure_count_lock = asyncio.Lock()
|
||||
|
||||
# 添加key失败计数记录
|
||||
key_failure_counts = {key: 0 for key in API_KEYS}
|
||||
MAX_FAILURES = 1 # 最大失败次数阈值
|
||||
MAX_FAILURES = 10 # 最大失败次数阈值
|
||||
MAX_RETRIES = 3 # 最大重试次数
|
||||
|
||||
|
||||
async def get_next_key():
|
||||
"""仅获取下一个key,不检查失败次数"""
|
||||
async with key_cycle_lock:
|
||||
return next(key_cycle)
|
||||
|
||||
async def is_key_valid(key):
|
||||
"""检查key是否有效"""
|
||||
async with failure_count_lock:
|
||||
return key_failure_counts[key] < MAX_FAILURES
|
||||
|
||||
async def reset_failure_counts():
|
||||
"""重置所有key的失败计数"""
|
||||
async with failure_count_lock:
|
||||
for key in key_failure_counts:
|
||||
key_failure_counts[key] = 0
|
||||
|
||||
async def get_next_working_key():
|
||||
"""获取下一个可用的API key"""
|
||||
async with key_lock:
|
||||
current_key = next(key_cycle)
|
||||
initial_key = current_key
|
||||
|
||||
while key_failure_counts[current_key] >= MAX_FAILURES:
|
||||
current_key = next(key_cycle)
|
||||
if current_key == initial_key: # 已经循环了一圈
|
||||
# 重置所有失败计数
|
||||
for key in key_failure_counts:
|
||||
key_failure_counts[key] = 0
|
||||
break
|
||||
|
||||
return current_key
|
||||
|
||||
initial_key = await get_next_key()
|
||||
current_key = initial_key
|
||||
|
||||
while True:
|
||||
if await is_key_valid(current_key):
|
||||
return current_key
|
||||
|
||||
current_key = await get_next_key()
|
||||
if current_key == initial_key: # 已经循环了一圈
|
||||
await reset_failure_counts()
|
||||
return current_key
|
||||
|
||||
async def handle_api_failure(api_key):
|
||||
"""处理API调用失败"""
|
||||
async with key_lock:
|
||||
async with failure_count_lock:
|
||||
key_failure_counts[api_key] += 1
|
||||
if key_failure_counts[api_key] >= MAX_FAILURES:
|
||||
logger.warning(
|
||||
f"API key {api_key} has failed {MAX_FAILURES} times, switching to next key"
|
||||
)
|
||||
return await get_next_working_key()
|
||||
return api_key
|
||||
logger.warning(f"API key {api_key} has failed {MAX_FAILURES} times, switching to next key")
|
||||
|
||||
# 在锁外获取新的key
|
||||
return await get_next_working_key()
|
||||
|
||||
|
||||
class ChatRequest(BaseModel):
|
||||
@@ -206,9 +222,8 @@ def convert_gemini_response_to_openai(response, model, stream=False):
|
||||
@app.get("/hf/v1/models")
|
||||
async def list_models(authorization: str = Header(None)):
|
||||
await verify_authorization(authorization)
|
||||
async with key_lock:
|
||||
api_key = next(key_cycle)
|
||||
logger.info(f"Using API key: {api_key}")
|
||||
api_key = await get_next_working_key()
|
||||
logger.info(f"Using API key: {api_key}")
|
||||
try:
|
||||
response = get_gemini_models(api_key)
|
||||
logger.info("Successfully retrieved models list")
|
||||
@@ -223,95 +238,124 @@ async def list_models(authorization: str = Header(None)):
|
||||
async def chat_completion(request: ChatRequest, authorization: str = Header(None)):
|
||||
await verify_authorization(authorization)
|
||||
api_key = await get_next_working_key()
|
||||
logger.info(f"Using API key: {api_key}")
|
||||
logger.info(f"Chat completion request - Model: {request.model}")
|
||||
retries = 0
|
||||
|
||||
while retries < MAX_RETRIES:
|
||||
try:
|
||||
logger.info(f"Attempt {retries + 1} with API key: {api_key}")
|
||||
|
||||
if request.model in config.settings.MODEL_SEARCH:
|
||||
# Gemini API调用部分
|
||||
gemini_messages = convert_messages_to_gemini_format(request.messages)
|
||||
# 调用Gemini API
|
||||
payload = {
|
||||
"contents": gemini_messages,
|
||||
"generationConfig": {
|
||||
"temperature": request.temperature,
|
||||
},
|
||||
"tools": [{"googleSearch": {}}],
|
||||
}
|
||||
|
||||
if request.stream:
|
||||
logger.info("Streaming response enabled")
|
||||
|
||||
try:
|
||||
logger.info(f"Chat completion request - Model: {request.model}")
|
||||
if request.model in config.settings.MODEL_SEARCH:
|
||||
# 转换消息格式
|
||||
gemini_messages = convert_messages_to_gemini_format(request.messages)
|
||||
|
||||
# 调用Gemini API
|
||||
non_stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:generateContent?key={api_key}"
|
||||
stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:streamGenerateContent?alt=sse&key={api_key}"
|
||||
payload = {
|
||||
"contents": gemini_messages,
|
||||
"generationConfig": {
|
||||
"temperature": request.temperature,
|
||||
},
|
||||
"tools": [{"googleSearch": {}}],
|
||||
}
|
||||
|
||||
if request.stream:
|
||||
async def generate():
|
||||
nonlocal api_key, retries
|
||||
while retries < MAX_RETRIES:
|
||||
try:
|
||||
async with httpx.AsyncClient() as client:
|
||||
stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:streamGenerateContent?alt=sse&key={api_key}"
|
||||
async with client.stream("POST", stream_url, json=payload) as response:
|
||||
if response.status_code == 429:
|
||||
logger.warning(f"Rate limit reached for key: {api_key}")
|
||||
api_key = await handle_api_failure(api_key)
|
||||
logger.info(f"Retrying with new API key: {api_key}")
|
||||
retries += 1
|
||||
if retries >= MAX_RETRIES:
|
||||
yield f"data: {json.dumps({'error': 'Max retries reached'})}\n\n"
|
||||
break
|
||||
continue
|
||||
|
||||
if response.status_code != 200:
|
||||
logger.error(f"Error in streaming response: {response.status_code}")
|
||||
yield f"data: {json.dumps({'error': f'API error: {response.status_code}'})}\n\n"
|
||||
break
|
||||
|
||||
async for line in response.aiter_lines():
|
||||
if line.startswith("data: "):
|
||||
try:
|
||||
chunk = json.loads(line[6:])
|
||||
openai_chunk = convert_gemini_response_to_openai(
|
||||
chunk, request.model, stream=True
|
||||
)
|
||||
if openai_chunk:
|
||||
yield f"data: {json.dumps(openai_chunk)}\n\n"
|
||||
except json.JSONDecodeError:
|
||||
continue
|
||||
yield "data: [DONE]\n\n"
|
||||
return
|
||||
except Exception as e:
|
||||
logger.error(f"Stream error: {str(e)}")
|
||||
api_key = await handle_api_failure(api_key)
|
||||
retries += 1
|
||||
if retries >= MAX_RETRIES:
|
||||
yield f"data: {json.dumps({'error': 'Max retries reached'})}\n\n"
|
||||
break
|
||||
continue
|
||||
|
||||
return StreamingResponse(content=generate(), media_type="text/event-stream")
|
||||
else:
|
||||
# 非流式响应
|
||||
async with httpx.AsyncClient() as client:
|
||||
non_stream_url = f"https://generativelanguage.googleapis.com/v1beta/models/{request.model}:generateContent?key={api_key}"
|
||||
response = await client.post(non_stream_url, json=payload)
|
||||
gemini_response = response.json()
|
||||
logger.info("Chat completion successful")
|
||||
return convert_gemini_response_to_openai(gemini_response, request.model)
|
||||
|
||||
# OpenAI API调用部分
|
||||
client = openai.OpenAI(api_key=api_key, base_url=config.settings.BASE_URL)
|
||||
response = client.chat.completions.create(
|
||||
model=request.model,
|
||||
messages=request.messages,
|
||||
temperature=request.temperature,
|
||||
stream=request.stream if hasattr(request, "stream") else False,
|
||||
)
|
||||
|
||||
if hasattr(request, "stream") and request.stream:
|
||||
logger.info("Streaming response enabled")
|
||||
|
||||
async def generate():
|
||||
async with httpx.AsyncClient() as client:
|
||||
async with client.stream(
|
||||
"POST", stream_url, json=payload
|
||||
) as response:
|
||||
async for line in response.aiter_lines():
|
||||
if line.startswith("data: "):
|
||||
try:
|
||||
chunk = json.loads(line[6:])
|
||||
openai_chunk = (
|
||||
convert_gemini_response_to_openai(
|
||||
chunk, request.model, stream=True
|
||||
)
|
||||
)
|
||||
if openai_chunk:
|
||||
yield f"data: {json.dumps(openai_chunk)}\n\n"
|
||||
except json.JSONDecodeError:
|
||||
continue
|
||||
yield "data: [DONE]\n\n"
|
||||
|
||||
return StreamingResponse(
|
||||
content=generate(), media_type="text/event-stream"
|
||||
)
|
||||
else:
|
||||
# 非流式响应
|
||||
async with httpx.AsyncClient() as client:
|
||||
response = await client.post(non_stream_url, json=payload)
|
||||
gemini_response = response.json()
|
||||
openai_response = convert_gemini_response_to_openai(
|
||||
gemini_response, request.model
|
||||
)
|
||||
|
||||
for chunk in response:
|
||||
yield f"data: {chunk.model_dump_json()}\n\n"
|
||||
logger.info("Chat completion successful")
|
||||
return openai_response
|
||||
client = openai.OpenAI(api_key=api_key, base_url=config.settings.BASE_URL)
|
||||
response = client.chat.completions.create(
|
||||
model=request.model,
|
||||
messages=request.messages,
|
||||
temperature=request.temperature,
|
||||
stream=request.stream if hasattr(request, "stream") else False,
|
||||
)
|
||||
return StreamingResponse(content=generate(), media_type="text/event-stream")
|
||||
|
||||
logger.info("Chat completion successful")
|
||||
return response
|
||||
|
||||
if hasattr(request, "stream") and request.stream:
|
||||
logger.info("Streaming response enabled")
|
||||
except Exception as e:
|
||||
logger.error(f"Error in chat completion: {str(e)}")
|
||||
api_key = await handle_api_failure(api_key)
|
||||
retries += 1
|
||||
|
||||
if retries >= MAX_RETRIES:
|
||||
logger.error("Max retries reached, giving up")
|
||||
raise HTTPException(status_code=500, detail="Max retries reached with all available API keys")
|
||||
|
||||
logger.info(f"Retrying with new API key: {api_key}")
|
||||
continue
|
||||
|
||||
async def generate():
|
||||
for chunk in response:
|
||||
yield f"data: {chunk.model_dump_json()}\n\n"
|
||||
|
||||
return StreamingResponse(content=generate(), media_type="text/event-stream")
|
||||
|
||||
logger.info("Chat completion successful")
|
||||
return response
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Error in chat completion: {str(e)}")
|
||||
api_key = await handle_api_failure(api_key) # 处理失败并可能切换key
|
||||
raise HTTPException(status_code=500, detail=str(e))
|
||||
raise HTTPException(status_code=500, detail="Unexpected error in chat completion")
|
||||
|
||||
|
||||
@app.post("/v1/embeddings")
|
||||
@app.post("/hf/v1/embeddings")
|
||||
async def embedding(request: EmbeddingRequest, authorization: str = Header(None)):
|
||||
await verify_authorization(authorization)
|
||||
async with key_lock:
|
||||
api_key = next(key_cycle)
|
||||
logger.info(f"Using API key: {api_key}")
|
||||
api_key = await get_next_working_key()
|
||||
logger.info(f"Using API key: {api_key}")
|
||||
|
||||
try:
|
||||
client = openai.OpenAI(api_key=api_key, base_url=config.settings.BASE_URL)
|
||||
|
||||
Reference in New Issue
Block a user