mle-agent 0.1.3

MLE-agent: An agent to automate your MLE processes

Keia: A Pair Agent for AI Engineer / Researchers

:love_letter: Fathers' love for Keia :love_letter:

Overview

The project is under active development. The API may change frequently.

Keia (MLE-Agent) is designed to be a pair agent for machine learning engineers or researchers. It includes two modes: Baseline Mode and Advanced Mode.

:coffee: Baseline Mode is designed to quickly build a baseline model for users' projects.

:fire: Advanced Mode (Coming Soon) is designed to utilize users' favorite MLOps tools, understand SOTA methods, and suggest optimizations for users' machine learning projects.

Milestones

:rocket: June 1st, 2024: Release the Baseline Mode (v0.1.0)

Get started

Installation

install from pypi

pip install mle-agent

install from source

git clone [e-mail]:MLSysOps/MLE-agent.git
pip install .

Configuration

You need to set up an LLM and choose tools before using the agent.

mle config

Usage (Baseline Mode)

Create a new project

mle new <project name>

Start a project

mle start

Other operations.

mle project ls # show all the available projects
mle project delete <project name> # delete a given project
mle project switch # switch the current working project

Roadmap

The following is a list of features that we plan to implement in the future. The list is not exhaustive, and we may add more features as we go along.

Plan, Generate, Execute and Debug Code

• An easy-to-use CLI interface
• Create/Select/Delete a project
• Understand users' requirements to suggest the file name, dataset, task, model arch, etc
• Generate a detailed coding plan
• Write baseline model code
• Execute the code on the local machine / cloud
• Debug the code and revise the code
• Googling the error message to debug the code
• Data Augmentation
• Hyperparameter tuning
• Model evaluation

Better user experience

• web interface (coming soon)
• discord bot

Integrate with AI/ML Tools

• snowflake / databricks
• wandb / mlflow
• skypilot
• dbt / airflow

Integrate with research tools

• huggingface
• paper with code
• arxiv

Contributing

We welcome contributions from the community. We are looking for contributors to help us with the following tasks:

• Benchmark and Evaluate the agent
• Add more features to the agent
• Improve the documentation
• Write tests

If you are interested in contributing, please check the CONTRIBUTING.md file.

Support and Community

• Discord community. If you have any questions, please feel free to ask in the Discord community.
• Twitter. Follow us on Twitter to get the latest updates.

Acknowledgements

We would like to thank the following contributors for their help with the project:

License

Check LICENSE file for more information.