phoenix/litellm-mirror
0
0
Fork 1
mirror of https://github.com/BerriAI/litellm.git synced 2025-04-27 19:54:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

(feat) add async, async+stream for gemini

Browse source
This commit is contained in:
ishaan-jaff 2023-12-16 18:58:12 +05:30
parent ea91cd7b83
commit 5a5ab37560
1 changed files with 40 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

44
litellm/llms/vertex_ai.py
View file

@ -229,8 +229,8 @@ def completion(
if acompletion == True: # [TODO] expand support to vertex ai chat + text models if acompletion == True: # [TODO] expand support to vertex ai chat + text models
if optional_params.get("stream", False) is True: if optional_params.get("stream", False) is True:
# async streaming # async streaming
return async_streaming(llm_model=llm_model, mode=mode, prompt=prompt, logging_obj=logging_obj, request_str=request_str, model=model, model_response=model_response, **optional_params) return async_streaming(llm_model=llm_model, mode=mode, prompt=prompt, logging_obj=logging_obj, request_str=request_str, model=model, model_response=model_response, messages=messages, print_verbose=print_verbose, **optional_params)
return async_completion(llm_model=llm_model, mode=mode, prompt=prompt, logging_obj=logging_obj, request_str=request_str, model=model, model_response=model_response, encoding=encoding, **optional_params) return async_completion(llm_model=llm_model, mode=mode, prompt=prompt, logging_obj=logging_obj, request_str=request_str, model=model, model_response=model_response, encoding=encoding, messages=messages,print_verbose=print_verbose,**optional_params)
if mode == "": if mode == "":
chat = llm_model.start_chat() chat = llm_model.start_chat()
@ -345,7 +345,7 @@ def completion(
except Exception as e: except Exception as e:
raise VertexAIError(status_code=500, message=str(e)) raise VertexAIError(status_code=500, message=str(e))
async def async_completion(llm_model, mode: str, prompt: str, model: str, model_response: ModelResponse, logging_obj=None, request_str=None, encoding=None, **optional_params): async def async_completion(llm_model, mode: str, prompt: str, model: str, model_response: ModelResponse, logging_obj=None, request_str=None, encoding=None, messages = None, print_verbose = None, **optional_params):
""" """
Add support for acompletion calls for gemini-pro Add support for acompletion calls for gemini-pro
""" """
@ -360,6 +360,24 @@ async def async_completion(llm_model, mode: str, prompt: str, model: str, model_
response_obj = await chat.send_message_async(prompt, generation_config=GenerationConfig(**optional_params)) response_obj = await chat.send_message_async(prompt, generation_config=GenerationConfig(**optional_params))
completion_response = response_obj.text completion_response = response_obj.text
response_obj = response_obj._raw_response response_obj = response_obj._raw_response
elif mode == "vision":
print_verbose("\nMaking VertexAI Gemini Pro Vision Call")
print_verbose(f"\nProcessing input messages = {messages}")
prompt, images = _gemini_vision_convert_messages(messages=messages)
content = [prompt] + images
request_str += f"response = llm_model.generate_content({content})\n"
## LOGGING
logging_obj.pre_call(input=prompt, api_key=None, additional_args={"complete_input_dict": optional_params, "request_str": request_str})
## LLM Call
response = await llm_model._generate_content_async(
contents=content,
generation_config=GenerationConfig(**optional_params)
)
completion_response = response.text
response_obj = response._raw_response
elif mode == "chat": elif mode == "chat":
# chat-bison etc. # chat-bison etc.
chat = llm_model.start_chat() chat = llm_model.start_chat()
@ -411,7 +429,7 @@ async def async_completion(llm_model, mode: str, prompt: str, model: str, model_
except Exception as e: except Exception as e:
raise VertexAIError(status_code=500, message=str(e)) raise VertexAIError(status_code=500, message=str(e))
async def async_streaming(llm_model, mode: str, prompt: str, model: str, model_response: ModelResponse, logging_obj=None, request_str=None, **optional_params): async def async_streaming(llm_model, mode: str, prompt: str, model: str, model_response: ModelResponse, logging_obj=None, request_str=None, messages = None, print_verbose = None, **optional_params):
""" """
Add support for async streaming calls for gemini-pro Add support for async streaming calls for gemini-pro
""" """
@ -425,6 +443,24 @@ async def async_streaming(llm_model, mode: str, prompt: str, model: str, model_r
logging_obj.pre_call(input=prompt, api_key=None, additional_args={"complete_input_dict": optional_params, "request_str": request_str}) logging_obj.pre_call(input=prompt, api_key=None, additional_args={"complete_input_dict": optional_params, "request_str": request_str})
response = await chat.send_message_async(prompt, generation_config=GenerationConfig(**optional_params), stream=stream) response = await chat.send_message_async(prompt, generation_config=GenerationConfig(**optional_params), stream=stream)
optional_params["stream"] = True optional_params["stream"] = True
elif mode == "vision":
stream = optional_params.pop("stream")
print_verbose("\nMaking VertexAI Gemini Pro Vision Call")
print_verbose(f"\nProcessing input messages = {messages}")
prompt, images = _gemini_vision_convert_messages(messages=messages)
content = [prompt] + images
stream = optional_params.pop("stream")
request_str += f"response = llm_model.generate_content({content}, generation_config=GenerationConfig(**{optional_params}), stream={stream})\n"
logging_obj.pre_call(input=prompt, api_key=None, additional_args={"complete_input_dict": optional_params, "request_str": request_str})
response = llm_model._generate_content_streaming_async(
contents=content,
generation_config=GenerationConfig(**optional_params),
stream=True
)
optional_params["stream"] = True
elif mode == "chat": elif mode == "chat":
chat = llm_model.start_chat() chat = llm_model.start_chat()
optional_params.pop("stream", None) # vertex ai raises an error when passing stream in optional params optional_params.pop("stream", None) # vertex ai raises an error when passing stream in optional params