forked from phoenix/litellm-mirror
feat(proxy_server.py): working /v1/messages with config.yaml
Adds async router support for adapter_completion call
This commit is contained in:
Krrish Dholakia
parent
2f8dbbeb97
commit
31829855c0
7 changed files with 362 additions and 14 deletions
|
|
@ -5045,23 +5045,187 @@ async def moderations(
|
|||
dependencies=[Depends(user_api_key_auth)],
|
||||
response_model=AnthropicResponse,
|
||||
)
|
||||
async def anthropic_response(data: AnthropicMessagesRequest):
|
||||
async def anthropic_response(
|
||||
anthropic_data: AnthropicMessagesRequest,
|
||||
fastapi_response: Response,
|
||||
request: Request,
|
||||
user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
|
||||
):
|
||||
from litellm import adapter_completion
|
||||
from litellm.adapters.anthropic_adapter import anthropic_adapter
|
||||
|
||||
litellm.adapters = [{"id": "anthropic", "adapter": anthropic_adapter}]
|
||||
|
||||
response: Optional[BaseModel] = adapter_completion(adapter_id="anthropic", **data)
|
||||
global user_temperature, user_request_timeout, user_max_tokens, user_api_base
|
||||
data: dict = {**anthropic_data, "adapter_id": "anthropic"}
|
||||
try:
|
||||
data["model"] = (
|
||||
general_settings.get("completion_model", None) # server default
|
||||
or user_model # model name passed via cli args
|
||||
or data["model"] # default passed in http request
|
||||
)
|
||||
if user_model:
|
||||
data["model"] = user_model
|
||||
|
||||
if response is None:
|
||||
raise Exception("Response is None.")
|
||||
elif not isinstance(response, AnthropicResponse):
|
||||
raise Exception(
|
||||
"Invalid model response={}. Not in 'AnthropicResponse' format".format(
|
||||
response
|
||||
data = await add_litellm_data_to_request(
|
||||
data=data, # type: ignore
|
||||
request=request,
|
||||
general_settings=general_settings,
|
||||
user_api_key_dict=user_api_key_dict,
|
||||
version=version,
|
||||
proxy_config=proxy_config,
|
||||
)
|
||||
|
||||
# override with user settings, these are params passed via cli
|
||||
if user_temperature:
|
||||
data["temperature"] = user_temperature
|
||||
if user_request_timeout:
|
||||
data["request_timeout"] = user_request_timeout
|
||||
if user_max_tokens:
|
||||
data["max_tokens"] = user_max_tokens
|
||||
if user_api_base:
|
||||
data["api_base"] = user_api_base
|
||||
|
||||
### MODEL ALIAS MAPPING ###
|
||||
# check if model name in model alias map
|
||||
# get the actual model name
|
||||
if data["model"] in litellm.model_alias_map:
|
||||
data["model"] = litellm.model_alias_map[data["model"]]
|
||||
|
||||
### CALL HOOKS ### - modify incoming data before calling the model
|
||||
data = await proxy_logging_obj.pre_call_hook( # type: ignore
|
||||
user_api_key_dict=user_api_key_dict, data=data, call_type="text_completion"
|
||||
)
|
||||
|
||||
### ROUTE THE REQUESTs ###
|
||||
router_model_names = llm_router.model_names if llm_router is not None else []
|
||||
# skip router if user passed their key
|
||||
if "api_key" in data:
|
||||
llm_response = asyncio.create_task(litellm.aadapter_completion(**data))
|
||||
elif (
|
||||
llm_router is not None and data["model"] in router_model_names
|
||||
): # model in router model list
|
||||
llm_response = asyncio.create_task(llm_router.aadapter_completion(**data))
|
||||
elif (
|
||||
llm_router is not None
|
||||
and llm_router.model_group_alias is not None
|
||||
and data["model"] in llm_router.model_group_alias
|
||||
): # model set in model_group_alias
|
||||
llm_response = asyncio.create_task(llm_router.aadapter_completion(**data))
|
||||
elif (
|
||||
llm_router is not None and data["model"] in llm_router.deployment_names
|
||||
): # model in router deployments, calling a specific deployment on the router
|
||||
llm_response = asyncio.create_task(
|
||||
llm_router.aadapter_completion(**data, specific_deployment=True)
|
||||
)
|
||||
elif (
|
||||
llm_router is not None and data["model"] in llm_router.get_model_ids()
|
||||
): # model in router model list
|
||||
llm_response = asyncio.create_task(llm_router.aadapter_completion(**data))
|
||||
elif (
|
||||
llm_router is not None
|
||||
and data["model"] not in router_model_names
|
||||
and llm_router.default_deployment is not None
|
||||
): # model in router deployments, calling a specific deployment on the router
|
||||
llm_response = asyncio.create_task(llm_router.aadapter_completion(**data))
|
||||
elif user_model is not None: # `litellm --model <your-model-name>`
|
||||
llm_response = asyncio.create_task(litellm.aadapter_completion(**data))
|
||||
else:
|
||||
raise HTTPException(
|
||||
status_code=status.HTTP_400_BAD_REQUEST,
|
||||
detail={
|
||||
"error": "completion: Invalid model name passed in model="
|
||||
+ data.get("model", "")
|
||||
},
|
||||
)
|
||||
|
||||
# Await the llm_response task
|
||||
response = await llm_response
|
||||
|
||||
hidden_params = getattr(response, "_hidden_params", {}) or {}
|
||||
model_id = hidden_params.get("model_id", None) or ""
|
||||
cache_key = hidden_params.get("cache_key", None) or ""
|
||||
api_base = hidden_params.get("api_base", None) or ""
|
||||
response_cost = hidden_params.get("response_cost", None) or ""
|
||||
|
||||
### ALERTING ###
|
||||
asyncio.create_task(
|
||||
proxy_logging_obj.update_request_status(
|
||||
litellm_call_id=data.get("litellm_call_id", ""), status="success"
|
||||
)
|
||||
)
|
||||
return response
|
||||
|
||||
verbose_proxy_logger.debug("final response: %s", response)
|
||||
|
||||
fastapi_response.headers.update(
|
||||
get_custom_headers(
|
||||
user_api_key_dict=user_api_key_dict,
|
||||
model_id=model_id,
|
||||
cache_key=cache_key,
|
||||
api_base=api_base,
|
||||
version=version,
|
||||
response_cost=response_cost,
|
||||
)
|
||||
)
|
||||
|
||||
verbose_proxy_logger.info("\nResponse from Litellm:\n{}".format(response))
|
||||
return response
|
||||
except RejectedRequestError as e:
|
||||
_data = e.request_data
|
||||
await proxy_logging_obj.post_call_failure_hook(
|
||||
user_api_key_dict=user_api_key_dict,
|
||||
original_exception=e,
|
||||
request_data=_data,
|
||||
)
|
||||
if _data.get("stream", None) is not None and _data["stream"] == True:
|
||||
_chat_response = litellm.ModelResponse()
|
||||
_usage = litellm.Usage(
|
||||
prompt_tokens=0,
|
||||
completion_tokens=0,
|
||||
total_tokens=0,
|
||||
)
|
||||
_chat_response.usage = _usage # type: ignore
|
||||
_chat_response.choices[0].message.content = e.message # type: ignore
|
||||
_iterator = litellm.utils.ModelResponseIterator(
|
||||
model_response=_chat_response, convert_to_delta=True
|
||||
)
|
||||
_streaming_response = litellm.TextCompletionStreamWrapper(
|
||||
completion_stream=_iterator,
|
||||
model=_data.get("model", ""),
|
||||
)
|
||||
|
||||
selected_data_generator = select_data_generator(
|
||||
response=_streaming_response,
|
||||
user_api_key_dict=user_api_key_dict,
|
||||
request_data=data,
|
||||
)
|
||||
|
||||
return StreamingResponse(
|
||||
selected_data_generator,
|
||||
media_type="text/event-stream",
|
||||
headers={},
|
||||
)
|
||||
else:
|
||||
_response = litellm.TextCompletionResponse()
|
||||
_response.choices[0].text = e.message
|
||||
return _response
|
||||
except Exception as e:
|
||||
await proxy_logging_obj.post_call_failure_hook(
|
||||
user_api_key_dict=user_api_key_dict, original_exception=e, request_data=data
|
||||
)
|
||||
verbose_proxy_logger.error(
|
||||
"litellm.proxy.proxy_server.completion(): Exception occured - {}".format(
|
||||
str(e)
|
||||
)
|
||||
)
|
||||
verbose_proxy_logger.debug(traceback.format_exc())
|
||||
error_msg = f"{str(e)}"
|
||||
raise ProxyException(
|
||||
message=getattr(e, "message", error_msg),
|
||||
type=getattr(e, "type", "None"),
|
||||
param=getattr(e, "param", "None"),
|
||||
code=getattr(e, "status_code", 500),
|
||||
)
|
||||
|
||||
|
||||
#### DEV UTILS ####
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue