ImportError: "Please install google-generativeai and 'vertexai' to use Google's API" Despite Both Libraries Being Installed.

[parth-sidgs]

What happened?

While running the script autogen_code_reviewer.py that utilizes the autogen library, the following error occurred:
ImportError: Please install google-generativeai and 'vertexai' to use Google's API.

This error persists even after installing both google-generativeai and vertexai in the virtual environment. I verified both libraries using the pip show command. Also, the script was activated in a fresh virtual environment with all packages installed. The same code was running 2 days back, but suddenly throwing the above error.

What did you expect to happen?

The script should have executed without throwing the ImportError, as the necessary dependencies (google-generativeai and vertexai) were installed successfully. Specifically:

1. The AssistantAgent initialization should proceed without issues.
2. The script should begin generating and reviewing code as expected.

How can we reproduce it (as minimally and precisely as possible)?

1. Activate the virtual environment.

2. Install the necessary dependencies:
   pip install pyautogen google-generativeai vertexai autogen-agentchat~=0.2

3. Run the following script:

import autogen
from autogen import AssistantAgent, UserProxyAgent

config_list = [
{
"model": "gemini-1.5-pro-002",
"api_type": "google",
"project_id": "my_project_id",
"location": "my_location"
}
]

code_generator = AssistantAgent(
name="code_generator",
llm_config={
"config_list": config_list,
"seed": 42,
"temperature": 0.7
},
system_message="""You are an expert code generator. Generate clean and well-documented code based on the requirements provided.""",
max_consecutive_auto_reply=2
)

Note: I authenticated with GCP using :gcloud auth application-default login, to use the Gemini model

AutoGen version

pyautogen version : 0.7.1

Which package was this bug in

Core

Model used

gemini-1.5-pro-002

Python version

3.12.5

Operating system

Windows 11

Any additional info you think would be helpful for fixing this bug

No response

[ekzhu]

pyautogen version : 0.7.1

We just released 0.4.3, where does the 0.7 come from? 😉

Please uninstall pyautogen and update your autogen-agentchat package to the latest version is v0.4.3, it's recommended to use that version.

Also, Gemini has OpenAI Compatible Endpoint, it should be available through OpenAIChatCompletionClient, see example. https://microsoft.github.io/autogen/dev/reference/python/autogen_ext.models.openai.html#autogen_ext.models.openai.OpenAIChatCompletionClient

For the use case, you can use a two-agent chat with RoundRobinGroupChat instead.

[parth-sidgs]

For pip show pyautogen-
it shows:
Name: pyautogen
Version: 0.7.1
Summary: A programming framework for agentic AI

Then i uninstalled pyautogen, and installed autogen (otherwise it throws module 'autogen' not found error)

Also upgraded autogen-agentchat, then did pip show autogen-agentchat -
output was:
Version: 0.4.3
Summary: AutoGen agents and teams library

Still getting the same error.

[ekzhu]

@parth-sidgs

A couple of things, first, we don't publish to pyautogen or autogen, please check the PyPI pages for details. Please uninstall both of them. They are conflicting with each other's and our import namespaces.

For your code, you are using the v0.2 API. You need to pin your autogen-agentchat~=0.2.0 to make sure you have the autogen.* module in your Python path. For v0.2 API:

import autogen
from autogen import AssistantAgent

config_list = [
{
"model": "gemini-1.5-flash",
"api_type": "openai",
"base_url": "https://generativelanguage.googleapis.com/v1beta/openai/",
}
]

code_generator = AssistantAgent(
name="code_generator",
llm_config={
"config_list": config_list,
"seed": 42,
"temperature": 0.7
},
system_message="""You are an expert code generator. Generate clean and well-documented code based on the requirements provided.""",
max_consecutive_auto_reply=2
)

We are using the OpenAI-compatible endpoint here. The native Gemini support in v0.2 is currently broken as you mentioned. We don't have the cycle to fix it at the moment. If you would like to contribute a fix would be great.

For v0.4 API, update your code to the following to use the v0.4 API autogen-agentchat~=0.4.0:

import asyncio
import os
from autogen_agentchat.agents import AssistantAgent
from autogen_ext.models.openai import OpenAIChatCompletionClient


async def main() -> None:
    model_client = OpenAIChatCompletionClient(
        model="gemini-1.5-flash",
        base_url="https://generativelanguage.googleapis.com/v1beta/openai/",
        api_key=os.environ["GEMINI_API_KEY"],
        model_info={
            "vision": True,
            "function_calling": True,
            "json_output": True,
            "family": "unknown",
        },
    )
    agent = AssistantAgent(
        "assistant",
        model_client=model_client,
    )
    print(await agent.run(task="Say 'Hello World!'"))


asyncio.run(main())

See Gemini's OpenAI Compatibility: https://ai.google.dev/gemini-api/docs/openai

Native Gemini client support in v0.4 is underway. Follow #5118

Migration guide: https://microsoft.github.io/autogen/stable/user-guide/agentchat-user-guide/migration-guide.html. Migrate to v0.4 is recommended.