Autogen? Autobuild!
AI Summary
Summary: Autobuild Feature in autogen
- Introduction:
- Presenter: Andy
- Topic: Autobuild feature in autogen
- Goal: Teach how to use Autobuild in projects
- Note: Andy is recovering from COVID-19
- Setting Up Environment:
- Use Visual Studio Code
- Create a new terminal window
- Create a Conda environment with Python 3.11.3
- Activate the new environment
- Install autogen with pip
- Create
oi_config_uncore_list
for API keys- Ensure the Conda environment is active in Visual Studio Code
- Using Autobuild:
- Import
AgentBuilder
from autogen- Set up configuration path and list
- Instantiate
AgentBuilder
with configuration and models- Set up a building task for the
AgentBuilder
- Execute
builder.build
with the task and coding flag set to trueAgentBuilder
returns agent list and configurations- Set up an execution task for the agents
- Create a group chat object and manager for agent interaction
- Initiate the chat with the first agent in the list
- Running the Project:
- Clear the terminal and run the project
- Review the output: agents generated and configurations prepared
- Agents create Python code but do not execute it
- Saving Configurations:
- Add
builder.save
to save agent configurations- Run the project again to generate and save new configurations
- Observations and Considerations:
- Agent creation for tasks is fairly consistent
- Agents may not always complete the given task
- User proxy agent lacks an LLM config, limiting its responses
- Cache settings may not be honored correctly
- Conclusion:
- Autobuild is powerful but requires careful prompting
- Links to project code and additional resources will be provided
- Resources:
- Project code gist
- Autobuild blog post
- Notebook the code was based on