[Feat]: support setting 'type' in tools for google search tool #14644
Description
What happened?
We are trying to test out google search grounding example using gemini-2.5-pro model.
{
"messages": [
{
"role": "system",
"content": "You are expert at cricket."
},
{
"role": "user",
"content": "Give me asia cup 2025 scores point table."
}
],
"tools": [
{
"type": "google_search",
"googleSearch": {}
}
]
}
As per litellm docs, the recommended way of testing is as shown below which is basically not passing tool type:
from litellm import completion
## SETUP ENVIRONMENT
# !gcloud auth application-default login - run this to add vertex credentials to your env
tools = [{"googleSearch": {}}] # 👈 ADD GOOGLE SEARCH
resp = litellm.completion(
model="vertex_ai/gemini-1.0-pro-001",
messages=[{"role": "user", "content": "Who won the world cup?"}],
tools=tools,
)
print(resp)
since we follow OPENAI spec, tool type is a required field and when we pass it in as shown above, we are seeing error:
BadRequestError: VertexAIException BadRequestError - {\n \"error\": {\n \"code\": 400,\n \"message\": \"The GenerateContentRequest proto is invalid:\\n * tools[0].tool_type: required one_of 'tool_type' must have one initialized field\",\n \"status\": \"INVALID_ARGUMENT\"\n }\n}\n"
Can somebody please take a look?
The transformed request going to vendor is basically not having googlesearch in the tools spec, hence its erroring out:
{'contents': [{'role': 'user', 'parts': [{'text': 'Give me asia cup 2025 scores point table.'}]}], 'system_instruction': {'parts': [{'text': 'You are expert at cricket.'}]}, 'tools': [{'function_declarations': []}], 'safetySettings': [{'category': 'HARM_CATEGORY_HARASSMENT', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_HATE_SPEECH', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_SEXUALLY_EXPLICIT', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_DANGEROUS_CONTENT', 'threshold': 'BLOCK_NONE'}], 'generationConfig': {}}
The correct transformation would look something like this:
{'contents': [{'role': 'user', 'parts': [{'text': 'Give me asia cup 2025 scores point table.'}]}], 'system_instruction': {'parts': [{'text': 'You are expert at cricket.'}]}, 'tools': [{'function_declarations': [], 'googleSearch': {}}], 'safetySettings': [{'category': 'HARM_CATEGORY_HARASSMENT', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_HATE_SPEECH', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_SEXUALLY_EXPLICIT', 'threshold': 'BLOCK_NONE'}, {'category': 'HARM_CATEGORY_DANGEROUS_CONTENT', 'threshold': 'BLOCK_NONE'}], 'generationConfig': {}}
Relevant log output
Shared aboveAre you a ML Ops Team?
Yes
What LiteLLM version are you on ?
v1.74.15.post2
Twitter / LinkedIn details
No response