Skip to content

README.md

Latest commit

 

History

History
84 lines (57 loc) · 2.81 KB

README.md

File metadata and controls

84 lines (57 loc) · 2.81 KB
Raw

ml-intern Introduction

What is ml-intern?

ml-intern is an open-source, autonomous AI agent developed by Hugging Face, designed to function as a specialized Machine Learning Engineer. It is built to handle the end-to-end lifecycle of ML tasks, from initial research to model deployment.

Core Philosophy

Specialized ML Engineering

Unlike general-purpose coding assistants, ml-intern is specifically optimized for the machine learning workflow. It understands the nuances of data processing, model training, and evaluation.

Deep Ecosystem Integration

ml-intern has native tools to interact with the Hugging Face Hub (datasets, models, spaces), documentation, and compute jobs, making it a powerful companion for researchers and developers in the HF ecosystem.

Autonomous Research

It can browse the web, search GitHub, and read academic papers (PDFs) to inform its coding decisions and stay up-to-date with the latest research.

Core Features

Autonomous Task Execution

  • Formulate hypotheses based on research papers.
  • Write high-quality ML code (training scripts, evaluation loops).
  • Run experiments and track results.
  • Deploy models and datasets directly to Hugging Face.

Advanced Agent Logic

  • Doom Loop Detector: Detects when the agent is stuck in repetitive loops and injects corrective prompts.
  • Auto-Compaction: Automatically manages long-running session contexts to keep the most relevant info in the LLM window.
  • Traceability: Sessions are auto-uploaded to a private HF dataset for debugging and auditing via the HF Agent Trace Viewer.

Multi-Model Support

  • Supports Claude (Anthropic), GPT (OpenAI), and local models (via Ollama, vLLM, or LM Studio).

Quick Start

Installation

Ensure you have uv installed.

# Clone the repository
git clone https://github.com/huggingface/ml-intern.git
cd ml-intern

# Sync dependencies
uv sync

# Install the CLI tool
uv tool install -e .

Configuration

Set up your environment variables (in .env or your shell):

export ANTHROPIC_API_KEY="your-key"
export OPENAI_API_KEY="your-key"
export HF_TOKEN="your-huggingface-token"
export GITHUB_TOKEN="your-github-token"

Basic Usage

# Start interactive mode
ml-intern

# Run a single task headlessly
ml-intern "fine-tune a ResNet model on the CIFAR-10 dataset"

# Specify a specific model
ml-intern --model anthropic/claude-5-fable "your prompt"

Related Resources

License

ml-intern is an open-source project by Hugging Face. Please follow their licensing terms and the terms of service of the LLM providers you use.