AutoTrain - AI Training Tools Tool

Overview

Hugging Face AutoTrain is an AutoML service and GUI that automates training, evaluation, and deployment of machine learning models across common tasks without requiring users to write training code. It supports a broad set of tasks — including text classification, token classification (NER), question answering, summarization, translation, image classification, tabular tasks, and LLM fine‑tuning — and is built to integrate tightly with the Hugging Face Hub for dataset and model management. AutoTrain handles dataset ingestion, preprocessing, automatic train/validation splits, model selection, and orchestration of training runs on hosted compute. AutoTrain is designed for both ML beginners and practitioners who want to accelerate iteration. Users can launch experiments from the web UI or via a client/CLI, configure few-shot or full-finetuning for large language models, run hyperparameter search and early stopping, and publish resulting models directly to the Hugging Face Hub for inference or further manual refinement. The platform emphasizes reproducibility (run metadata and logs), experiment comparison, and one-click deployment using Hugging Face Inference endpoints, making it suitable for prototyping and production workflows.

Key Features

• No-code web UI for dataset upload, labeling, and experiment launch
• Supports text classification, token classification, QA, summarization, translation
• Image classification and tabular task workflows with automatic preprocessing
• LLM fine‑tuning workflows, including instruction‑tuning and adapter support
• Automatic train/validation splitting, preprocessing, and label inference
• Built‑in hyperparameter search, early stopping, and run comparison UI
• Seamless Hub integration: publish models and deploy via Inference API

Example Usage

Example (python):

from autotrain import AutoTrain

# Authenticate with your Hugging Face token (set HF_TOKEN in your environment)
client = AutoTrain(api_token="$HF_TOKEN")

# Create a new AutoTrain project for text classification
project = client.create_project(name="sentiment-demo", task="text_classification")

# Upload a CSV/JSONL dataset (columns inferred automatically)
project.upload_file(path="data/train.csv", split="train")
project.upload_file(path="data/valid.csv", split="validation")

# Start a training run with default model search and metric
run = project.create_run(display_name="distilbert-autotrain-run", metric="accuracy")

# Wait for completion and fetch resulting model information
run.wait_for_completion()
model_info = run.get_model()

# Publish trained model to the Hugging Face Hub
model_info.push_to_hub(repo_id="your-username/sentiment-autotrain")

print("Trained model pushed to Hub:", model_info.repo_id)

Key Information

• Category: Training Tools
• Type: AI Training Tools Tool