Aim
ActiveDescription
Aim is an open-source experiment tracker and AI metadata store with a comparison-friendly web UI, rich visualizations, and a queryable API for ML and LLM observability.
Key Features
- Training and AI metadata logging - simple API to log runs, hyperparameters, metrics, and artifacts
- Comparison-friendly UI - powerful web UI to compare hundreds of runs side by side
- Rich visualizations - Plotly-style charts, images, audio, text, and other modalities
- Programmable query API - SDK and CLI for programmatic access to experiment data
- Wide framework integration - works with PyTorch Lightning, Hugging Face, TensorFlow, and more
- Self-hostable and extensible - fully open-source with pluggable storage backends
Use Cases
💡 Track and compare LLM fine-tuning experiments and evaluation metrics
💡 Share training results and visualizations across teams
💡 Pipe experiment metadata into custom dashboards via the API
💡 Debug loss curves and gradients for long-running training jobs
💡 Log and replay model evaluation runs from CI
Categories
Quick Start
# Install Aim
pip install aim
# Log experiments from your training script
from aim import Run
run = Run()
run["hparams"] = {"lr": 0.001, "batch_size": 32}
for step in range(100):
run.track(0.9 ** step, name="loss", step=step)
# Launch the web UI to explore runs
aim up