wtf is ml-intern?

ML Intern is a command-line tool that acts as an AI assistant specifically for machine learning work. You give it a task in plain English, such as "fine-tune a language model on my dataset," and it figures out the steps, writes code, and executes them. It has built-in access to Hugging Face's ecosystem of tools, including research papers, open datasets, documentation, and cloud computing resources. The tool runs in two modes. Interactive mode opens a chat session where you type requests and the agent responds and takes action. Headless mode lets you pass a single instruction as a command and walk away. You can point it at different AI models to power the reasoning, including models from Anthropic, OpenAI, locally running models via Ollama or vLLM, and various models available through the Hugging Face router. By default, the tool runs code directly on your local machine, reading and writing files in your project folder. If you need GPU resources or want to test code in an isolated environment, you can switch to sandbox mode, which creates a private cloud workspace on Hugging Face Spaces. This is useful for running training jobs that would be too slow or impractical on a laptop. Every session is automatically saved to a private dataset on your Hugging Face account, so you can review what the agent did, what tools it called, and how the model responded. You can make these traces public or keep them private. There is also optional Slack integration that sends you status updates when the agent needs approval or finishes a task, which is handy if you kick off a long job and step away. The project is open source under the Apache 2.0 license and is maintained by the Hugging Face team.