fix: fix the issues when using tools in gemini

[kan-bayashi]

This PR fixes the following two issues:

• Error when using multiple tools
• Unexpected behavior when response.candidates[0].content.parts[0].function_call exists but empty (not None)

Error reproducible code

import litellm
tools = [
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather in a given location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g. San Francisco, CA",
                    },
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "get_current_stock_price",
            "description": "Get the current stock price",
            "parameters": {
                "type": "object",
                "properties": {},
                "required": [],
            },
        },
    },
]
messages = [{"role": "user", "content": "What's the weather like in Boston today?"}]
completion = litellm.completion(
    model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
)
print(messages)
print(completion)
messages = [{"role": "user", "content": "Please tell me the stock price."}]
completion = litellm.completion(
    model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
)
print(messages)
print(completion)

Before

Give Feedback / Get Help: https://github.com/BerriAI/litellm/issues/new
LiteLLM.Info: If you need to debug this error, use `litellm.set_verbose=True'.

---------------------------------------------------------------------------
_InactiveRpcError                         Traceback (most recent call last)
File ~/test-litellm/.venv/lib/python3.10/site-packages/google/api_core/grpc_helpers.py:79, in _wrap_unary_errors.<locals>.error_remapped_callable(*args, **kwargs)
     78 try:
---> 79     return callable_(*args, **kwargs)
     80 except grpc.RpcError as exc:

File ~/test-litellm/.venv/lib/python3.10/site-packages/grpc/_channel.py:1160, in _UnaryUnaryMultiCallable.__call__(self, request, timeout, metadata, credentials, wait_for_ready, compression)
   1154 (
   1155     state,
   1156     call,
   1157 ) = self._blocking(
   1158     request, timeout, metadata, credentials, wait_for_ready, compression
   1159 )
-> 1160 return _end_unary_response_blocking(state, call, False, None)

File ~/test-litellm/.venv/lib/python3.10/site-packages/grpc/_channel.py:1003, in _end_unary_response_blocking(state, call, with_call, deadline)
   1002 else:
-> 1003     raise _InactiveRpcError(state)

_InactiveRpcError: <_InactiveRpcError of RPC that terminated with:
        status = StatusCode.INVALID_ARGUMENT
        details = "At most one tool is supported."
        debug_error_string = "UNKNOWN:Error received from peer ipv4:172.217.25.170:443 {grpc_message:"At most one tool is supported.", grpc_status:3, created_time:"2024-02-14T13:07:52.970947+09:00"}"
>

The above exception was the direct cause of the following exception:

InvalidArgument                           Traceback (most recent call last)
File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/llms/vertex_ai.py:428, in completion(model, messages, model_response, print_verbose, encoding, logging_obj, vertex_project, vertex_location, optional_params, litellm_params, logger_fn, acompletion)
    427 ## LLM Call
--> 428 response = llm_model.generate_content(
    429     contents=content,
    430     generation_config=GenerationConfig(**optional_params),
    431     safety_settings=safety_settings,
    432     tools=tools,
    433 )
    435 if tools is not None and hasattr(
    436     response.candidates[0].content.parts[0], "function_call"
    437 ):

File ~/test-litellm/.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:353, in _GenerativeModel.generate_content(self, contents, generation_config, safety_settings, tools, stream)
    352 else:
--> 353     return self._generate_content(
    354         contents=contents,
    355         generation_config=generation_config,
    356         safety_settings=safety_settings,
    357         tools=tools,
    358     )

File ~/test-litellm/.venv/lib/python3.10/site-packages/vertexai/generative_models/_generative_models.py:434, in _GenerativeModel._generate_content(self, contents, generation_config, safety_settings, tools)
    428 request = self._prepare_request(
    429     contents=contents,
    430     generation_config=generation_config,
    431     safety_settings=safety_settings,
    432     tools=tools,
    433 )
--> 434 gapic_response = self._prediction_client.generate_content(request=request)
    435 return self._parse_response(gapic_response)

File ~/test-litellm/.venv/lib/python3.10/site-packages/google/cloud/aiplatform_v1beta1/services/prediction_service/client.py:2075, in PredictionServiceClient.generate_content(self, request, model, contents, retry, timeout, metadata)
   2074 # Send the request.
-> 2075 response = rpc(
   2076     request,
   2077     retry=retry,
   2078     timeout=timeout,
   2079     metadata=metadata,
   2080 )
   2082 # Done; return the response.

File ~/test-litellm/.venv/lib/python3.10/site-packages/google/api_core/gapic_v1/method.py:131, in _GapicCallable.__call__(self, timeout, retry, compression, *args, **kwargs)
    129     kwargs["compression"] = compression
--> 131 return wrapped_func(*args, **kwargs)

File ~/test-litellm/.venv/lib/python3.10/site-packages/google/api_core/grpc_helpers.py:81, in _wrap_unary_errors.<locals>.error_remapped_callable(*args, **kwargs)
     80 except grpc.RpcError as exc:
---> 81     raise exceptions.from_grpc_error(exc) from exc

InvalidArgument: 400 At most one tool is supported.

During handling of the above exception, another exception occurred:

VertexAIError                             Traceback (most recent call last)
File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/main.py:1472, in completion(model, messages, timeout, temperature, top_p, n, stream, stop, max_tokens, presence_penalty, frequency_penalty, logit_bias, user, response_format, seed, tools, tool_choice, logprobs, top_logprobs, deployment_id, functions, function_call, base_url, api_version, api_key, model_list, **kwargs)
   1466 vertex_ai_location = (
   1467     optional_params.pop("vertex_ai_location", None)
   1468     or litellm.vertex_location
   1469     or get_secret("VERTEXAI_LOCATION")
   1470 )
-> 1472 model_response = vertex_ai.completion(
   1473     model=model,
   1474     messages=messages,
   1475     model_response=model_response,
   1476     print_verbose=print_verbose,
   1477     optional_params=optional_params,
   1478     litellm_params=litellm_params,
   1479     logger_fn=logger_fn,
   1480     encoding=encoding,
   1481     vertex_location=vertex_ai_location,
   1482     vertex_project=vertex_ai_project,
   1483     logging_obj=logging,
   1484     acompletion=acompletion,
   1485 )
   1487 if (
   1488     "stream" in optional_params
   1489     and optional_params["stream"] == True
   1490     and acompletion == False
   1491 ):

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/llms/vertex_ai.py:614, in completion(model, messages, model_response, print_verbose, encoding, logging_obj, vertex_project, vertex_location, optional_params, litellm_params, logger_fn, acompletion)
    613 except Exception as e:
--> 614     raise VertexAIError(status_code=500, message=str(e))

VertexAIError: 400 At most one tool is supported.

During handling of the above exception, another exception occurred:

RateLimitError                            Traceback (most recent call last)
Cell In[1], line 35
      2 tools = [
      3     {
      4         "type": "function",
   (...)
     32     },
     33 ]
     34 messages = [{"role": "user", "content": "What's the weather like in Boston today?"}]
---> 35 completion = litellm.completion(
     36     model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
     37 )

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/utils.py:2474, in client.<locals>.wrapper(*args, **kwargs)
   2470         if (
   2471             liteDebuggerClient and liteDebuggerClient.dashboard_url != None
   2472         ):  # make it easy to get to the debugger logs if you've initialized it
   2473             e.message += f"\n Check the log in your dashboard - {liteDebuggerClient.dashboard_url}"
-> 2474 raise e

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/utils.py:2377, in client.<locals>.wrapper(*args, **kwargs)
   2375         print_verbose(f"Error while checking max token limit: {str(e)}")
   2376 # MODEL CALL
-> 2377 result = original_function(*args, **kwargs)
   2378 end_time = datetime.datetime.now()
   2379 if "stream" in kwargs and kwargs["stream"] == True:

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/main.py:1898, in completion(model, messages, timeout, temperature, top_p, n, stream, stop, max_tokens, presence_penalty, frequency_penalty, logit_bias, user, response_format, seed, tools, tool_choice, logprobs, top_logprobs, deployment_id, functions, function_call, base_url, api_version, api_key, model_list, **kwargs)
   1895     return response
   1896 except Exception as e:
   1897     ## Map to OpenAI Exception
-> 1898     raise exception_type(
   1899         model=model,
   1900         custom_llm_provider=custom_llm_provider,
   1901         original_exception=e,
   1902         completion_kwargs=args,
   1903     )

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/utils.py:7513, in exception_type(model, original_exception, custom_llm_provider, completion_kwargs)
   7511 # don't let an error with mapping interrupt the user from receiving an error from the llm api calls
   7512 if exception_mapping_worked:
-> 7513     raise e
   7514 else:
   7515     raise original_exception

File ~/test-litellm/.venv/lib/python3.10/site-packages/litellm/utils.py:6765, in exception_type(model, original_exception, custom_llm_provider, completion_kwargs)
   6760 elif (
   6761     "429 Quota exceeded" in error_str
   6762     or "IndexError: list index out of range"
   6763 ):
   6764     exception_mapping_worked = True
-> 6765     raise RateLimitError(
   6766         message=f"VertexAIException - {error_str}",
   6767         model=model,
   6768         llm_provider="vertex_ai",
   6769         response=httpx.Response(
   6770             status_code=429,
   6771             request=httpx.Request(
   6772                 method="POST",
   6773                 url=" https://cloud.google.com/vertex-ai/",
   6774             ),
   6775         ),
   6776     )
   6777 if hasattr(original_exception, "status_code"):
   6778     if original_exception.status_code == 400:

RateLimitError: VertexAIException - 400 At most one tool is supported.

After fix the first issue

[{'role': 'user', 'content': "What's the weather like in Boston today?"}]
ModelResponse(id='chatcmpl-050a4c91-f559-4af8-8e83-09ba2fd56132', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='Message(content=None, role=\'assistant\', tool_calls=[ChatCompletionMessageToolCall(id=\'call_2794dccb-6346-4d3e-97ba-12bcf78fbf75\', function=Function(arguments=\'{"location": "Boston, MA"}\', name=\'get_current_weather\'), type=\'function\')])', role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_2794dccb-6346-4d3e-97ba-12bcf78fbf75', function=Function(arguments='{"location": "Boston, MA"}', name='get_current_weather'), type='function')]))], created=1707889900, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=55, completion_tokens=9, total_tokens=64))
[{'role': 'user', 'content': 'Please tell me the stock price.'}]
ModelResponse(id='chatcmpl-037e7e96-acef-4d79-b3f8-9a15580da6dc', choices=[Choices(finish_reason='STOP', index=0, message=Message(content="Message(content=None, role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_e30ca22b-3eb5-4f3e-8152-c262ec088beb', function=Function(arguments='{}', name=''), type='function')])", role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_e30ca22b-3eb5-4f3e-8152-c262ec088beb', function=Function(arguments='{}', name=''), type='function')]))], created=1707889902, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=52, completion_tokens=18, total_tokens=70))

After fix both issues

[{'role': 'user', 'content': "What's the weather like in Boston today?"}]
ModelResponse(id='chatcmpl-5c7f5560-6736-4654-83ad-58f7a7ea86eb', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='Message(content=None, role=\'assistant\', tool_calls=[ChatCompletionMessageToolCall(id=\'call_32580eac-7ce5-411d-a30b-e622d2af38ad\', function=Function(arguments=\'{"location": "Boston, MA"}\', name=\'get_current_weather\'), type=\'function\')])', role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_32580eac-7ce5-411d-a30b-e622d2af38ad', function=Function(arguments='{"location": "Boston, MA"}', name='get_current_weather'), type='function')]))], created=1707899384, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=55, completion_tokens=9, total_tokens=64))
[{'role': 'user', 'content': 'Please tell me the stock price.'}]
ModelResponse(id='chatcmpl-6f8ea342-cad2-4d95-a7b3-777b026ea799', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='I am sorry, I cannot fulfill this request. The available tools lack the desired functionality.', role='assistant'))], created=1707899385, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=52, completion_tokens=18, total_tokens=70))

[vercel]

The latest updates on your projects. Learn more about Vercel for Git ↗︎

Name	Status	Preview	Comments	Updated (UTC)
litellm	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 14, 2024 8:40am
litellm-dashboard	✅ Ready (Inspect)	Visit Preview	💬 Add feedback	Feb 14, 2024 8:40am

[krrishdholakia]

@kan-bayashi could you upload a screenshot of this working with multiple tools for confirmation?

[kan-bayashi]

Hi @krrishdholakia. I updated the PR description, please check it.
During testing, I found the another issue when using tool without argument.
(See the following error)

Give Feedback / Get Help: https://github.com/BerriAI/litellm/issues/new
LiteLLM.Info: If you need to debug this error, use `litellm.set_verbose=True'.

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
File ~/work/litellm/litellm/llms/vertex_ai.py:440, in completion(model, messages, model_response, print_verbose, encoding, logging_obj, vertex_project, vertex_location, optional_params, litellm_params, logger_fn, acompletion)
    439 args_dict = {}
--> 440 for k, v in function_call.args.items():
    441     args_dict[k] = v

AttributeError: 'NoneType' object has no attribute 'items'

During handling of the above exception, another exception occurred:

VertexAIError                             Traceback (most recent call last)
File ~/work/litellm/litellm/main.py:1472, in completion(model, messages, timeout, temperature, top_p, n, stream, stop, max_tokens, presence_penalty, frequency_penalty, logit_bias, user, response_format, seed, tools, tool_choice, logprobs, top_logprobs, deployment_id, functions, function_call, base_url, api_version, api_key, model_list, **kwargs)
   1466 vertex_ai_location = (
   1467     optional_params.pop("vertex_ai_location", None)
   1468     or litellm.vertex_location
   1469     or get_secret("VERTEXAI_LOCATION")
   1470 )
-> 1472 model_response = vertex_ai.completion(
   1473     model=model,
   1474     messages=messages,
   1475     model_response=model_response,
   1476     print_verbose=print_verbose,
   1477     optional_params=optional_params,
   1478     litellm_params=litellm_params,
   1479     logger_fn=logger_fn,
   1480     encoding=encoding,
   1481     vertex_location=vertex_ai_location,
   1482     vertex_project=vertex_ai_project,
   1483     logging_obj=logging,
   1484     acompletion=acompletion,
   1485 )
   1487 if (
   1488     "stream" in optional_params
   1489     and optional_params["stream"] == True
   1490     and acompletion == False
   1491 ):

File ~/work/litellm/litellm/llms/vertex_ai.py:614, in completion(model, messages, model_response, print_verbose, encoding, logging_obj, vertex_project, vertex_location, optional_params, litellm_params, logger_fn, acompletion)
    613 except Exception as e:
--> 614     raise VertexAIError(status_code=500, message=str(e))

VertexAIError: 'NoneType' object has no attribute 'items'

During handling of the above exception, another exception occurred:

RateLimitError                            Traceback (most recent call last)
Cell In[2], line 41
     39 print(completion)
     40 messages = [{"role": "user", "content": "Please tell me the stock price."}]
---> 41 completion = litellm.completion(
     42     model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
     43 )
     44 print(completion)

File ~/work/litellm/litellm/utils.py:2474, in client.<locals>.wrapper(*args, **kwargs)
   2470         if (
   2471             liteDebuggerClient and liteDebuggerClient.dashboard_url != None
   2472         ):  # make it easy to get to the debugger logs if you've initialized it
   2473             e.message += f"\n Check the log in your dashboard - {liteDebuggerClient.dashboard_url}"
-> 2474 raise e

File ~/work/litellm/litellm/utils.py:2377, in client.<locals>.wrapper(*args, **kwargs)
   2375         print_verbose(f"Error while checking max token limit: {str(e)}")
   2376 # MODEL CALL
-> 2377 result = original_function(*args, **kwargs)
   2378 end_time = datetime.datetime.now()
   2379 if "stream" in kwargs and kwargs["stream"] == True:

File ~/work/litellm/litellm/main.py:1898, in completion(model, messages, timeout, temperature, top_p, n, stream, stop, max_tokens, presence_penalty, frequency_penalty, logit_bias, user, response_format, seed, tools, tool_choice, logprobs, top_logprobs, deployment_id, functions, function_call, base_url, api_version, api_key, model_list, **kwargs)
   1895     return response
   1896 except Exception as e:
   1897     ## Map to OpenAI Exception
-> 1898     raise exception_type(
   1899         model=model,
   1900         custom_llm_provider=custom_llm_provider,
   1901         original_exception=e,
   1902         completion_kwargs=args,
   1903     )

File ~/work/litellm/litellm/utils.py:7510, in exception_type(model, original_exception, custom_llm_provider, completion_kwargs)
   7508 # don't let an error with mapping interrupt the user from receiving an error from the llm api calls
   7509 if exception_mapping_worked:
-> 7510     raise e
   7511 else:
   7512     raise original_exception

File ~/work/litellm/litellm/utils.py:6762, in exception_type(model, original_exception, custom_llm_provider, completion_kwargs)
   6757 elif (
   6758     "429 Quota exceeded" in error_str
   6759     or "IndexError: list index out of range"
   6760 ):
   6761     exception_mapping_worked = True
-> 6762     raise RateLimitError(
   6763         message=f"VertexAIException - {error_str}",
   6764         model=model,
   6765         llm_provider="vertex_ai",
   6766         response=httpx.Response(
   6767             status_code=429,
   6768             request=httpx.Request(
   6769                 method="POST",
   6770                 url=" https://cloud.google.com/vertex-ai/",
   6771             ),
   6772         ),
   6773     )
   6774 if hasattr(original_exception, "status_code"):
   6775     if original_exception.status_code == 400:

RateLimitError: VertexAIException - 'NoneType' object has no attribute 'items'

I also fixed the issue in 917525e (#1969).

[kan-bayashi]

@krrishdholakia
I found the second completion in the above is failed due to the wrong LLM outputs.
(function_call = response.candidates[0].content.parts[0].function_call exists but empty)
Therefore, I think the following part

litellm/litellm/llms/vertex_ai.py

Lines 435 to 437 in e590f47

	if tools is not None and hasattr(
	response.candidates[0].content.parts[0], "function_call"
	):

should be

            if tools is not None and bool(
                getattr(response.candidates[0].content.parts[0], "function_call", None)
            ):

Then second completion looks fine.

# Before
[{'role': 'user', 'content': 'Please tell me the stock price.'}]
ModelResponse(id='chatcmpl-037e7e96-acef-4d79-b3f8-9a15580da6dc', choices=[Choices(finish_reason='STOP', index=0, message=Message(content="Message(content=None, role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_e30ca22b-3eb5-4f3e-8152-c262ec088beb', function=Function(arguments='{}', name=''), type='function')])", role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_e30ca22b-3eb5-4f3e-8152-c262ec088beb', function=Function(arguments='{}', name=''), type='function')]))], created=1707889902, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=52, completion_tokens=18, total_tokens=70))

# After
[{'role': 'user', 'content': 'Please tell me the stock price.'}]
ModelResponse(id='chatcmpl-6f8ea342-cad2-4d95-a7b3-777b026ea799', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='I am sorry, I cannot fulfill this request. The available tools lack the desired functionality.', role='assistant'))], created=1707899385, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=52, completion_tokens=18, total_tokens=70))

I will make a commit to fix this issue.

[kan-bayashi]

Now everything works fine! Please let me know if you find wondering points.

[kan-bayashi]

Another example works correctly with multiple tools.

import litellm

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_current_weather",
            "description": "Get the current weather in a given location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g. San Francisco, CA",
                    },
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "get_tomorrow_weather",
            "description": "Get the tomorrow weather in a given location",
            "parameters": {
                "type": "object",
                "properties": {
                    "location": {
                        "type": "string",
                        "description": "The city and state, e.g. San Francisco, CA",
                    },
                    "unit": {"type": "string", "enum": ["celsius", "fahrenheit"]},
                },
                "required": ["location"],
            },
        },
    },
    {
        "type": "function",
        "function": {
            "name": "get_yesterday_newyork_weather",
            "description": "Get the tomorrow weather in New York yesterday",
            "parameters": {
                "type": "object",
                "properties": {},
                "required": [],
            },
        },
    },
]
messages = [{"role": "user", "content": "What's the weather like in Boston today?"}]
completion = litellm.completion(
    model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
)
print(messages)
print(completion)

messages = [{"role": "user", "content": "What's the weather like in Boston tomorrow?"}]
completion = litellm.completion(
    model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
)
print(messages)
print(completion)

messages = [{"role": "user", "content": "What's the weather like in New York yesterday?"}]
completion = litellm.completion(
    model="gemini-pro", messages=messages, tools=tools, tool_choice="auto"
)
print(messages)
print(completion)

[{'role': 'user', 'content': "What's the weather like in Boston today?"}]
ModelResponse(id='chatcmpl-69d45e30-6b17-422f-8cf5-8c0c802e52b7', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='Message(content=None, role=\'assistant\', tool_calls=[ChatCompletionMessageToolCall(id=\'call_78d33abc-b931-4eb7-b01f-14e58e1949c2\', function=Function(arguments=\'{"location": "Boston, MA"}\', name=\'get_current_weather\'), type=\'function\')])', role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_78d33abc-b931-4eb7-b01f-14e58e1949c2', function=Function(arguments='{"location": "Boston, MA"}', name='get_current_weather'), type='function')]))], created=1707955914, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=92, completion_tokens=9, total_tokens=101))
[{'role': 'user', 'content': "What's the weather like in Boston tomorrow?"}]
ModelResponse(id='chatcmpl-09868eb0-b20c-4513-a354-fa7b25dcff2e', choices=[Choices(finish_reason='STOP', index=0, message=Message(content='Message(content=None, role=\'assistant\', tool_calls=[ChatCompletionMessageToolCall(id=\'call_96b762d1-81ff-49d1-86cc-8072c269ab48\', function=Function(arguments=\'{"location": "Boston, MA"}\', name=\'get_tomorrow_weather\'), type=\'function\')])', role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_96b762d1-81ff-49d1-86cc-8072c269ab48', function=Function(arguments='{"location": "Boston, MA"}', name='get_tomorrow_weather'), type='function')]))], created=1707955921, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=92, completion_tokens=9, total_tokens=101))
[{'role': 'user', 'content': "What's the weather like in New York yesterday?"}]
ModelResponse(id='chatcmpl-21ede8d8-ec50-46fe-8b01-3ab1c02900c2', choices=[Choices(finish_reason='STOP', index=0, message=Message(content="Message(content=None, role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_94806bb6-0e1c-4956-bac7-e0c10b428c3f', function=Function(arguments='{}', name='get_yesterday_newyork_weather'), type='function')])", role='assistant', tool_calls=[ChatCompletionMessageToolCall(id='call_94806bb6-0e1c-4956-bac7-e0c10b428c3f', function=Function(arguments='{}', name='get_yesterday_newyork_weather'), type='function')]))], created=1707955928, model='gemini-pro', object='chat.completion', system_fingerprint=None, usage=Usage(prompt_tokens=93, completion_tokens=8, total_tokens=101))

[themrzmaster]

nice! i need that fix