aimhubio/aim

Features

• Experiment Tracking - Logs hyperparameters, metrics, and artifacts from training runs to compare and visualize model performance.
• Machine Learning Experiment Trackers - An open-source platform for logging, visualizing, and comparing machine learning training runs and their metadata.
• Artifact Logging - Logs arbitrary files and objects produced during a run and associates them with the experiment for retrieval.
• Code-Integrated Training Frameworks - Connects to popular ML frameworks like PyTorch and TensorFlow for logging training runs with minimal code changes.
• Experiment Logging - Records hyperparameters, metrics, and artifacts across training runs and visualizes them in an interactive UI.
• Experiment Tracking Servers - Provides a self-hosted server that centralizes experiment data collection from multiple clients for collaborative analysis.
• LLM Execution Tracing - Captures prompts, generations, and tool calls from LLM workflows for debugging and comparison.
• ML Library Integrations - Provides a Python SDK that integrates with PyTorch, TensorFlow, Hugging Face, and XGBoost with minimal code changes.
• Training Log Analysis - Records metrics, parameters, and metadata from machine learning training runs for later analysis and comparison.
• Experiment Run Explorers - Uses specialized explorers to compare thousands of sessions of metrics, images, text, and audio.
• Run Comparison Tools - Provides a web UI that displays multiple training runs side by side to highlight performance differences.
• Interactive Dashboards - Ships an interactive dashboard that displays side-by-side views of training runs to highlight performance differences.
• Run Data Models - Stores each training run as a self-contained record of metrics, parameters, and artifacts for independent comparison.
• LLM Execution Tracing - Logs prompts, generations, and tool calls from LLM applications for debugging and comparison.
• Training Metric Streaming - Streams scalar metrics from training processes to the UI via WebSocket or polling for live chart updates.
• Training Metric Loggers - Plots real-time charts of metrics such as loss and accuracy as the model trains.
• Experiment Tracking Servers - Runs a dedicated server that accepts experiment data from remote clients for centralized tracking.
• Experiment Run Comparators - Visualizes and compares thousands of experiment runs, metrics, images, and text in an interactive UI for pattern discovery.
• Logging Systems - Ships a configurable logging system that captures metrics, media, and LLM traces across AI pipelines.
• Experiment Tracking Servers - Exposes a remote HTTP API for logging and querying training runs from multiple clients.
• Local SQLite Stores - Persists experiment metadata and run data in a local SQLite database for fast, offline-accessible retrieval.
• Programmatic Query Languages - Provides a structured query language for filtering and retrieving experiment data programmatically.
• Media and Object Logging - Logs images, audio, distributions, and text alongside scalar metrics for richer experiment analysis.
• Pipeline Stage Tracing - Captures the full sequence of retrieval, context assembly, and generation steps to debug RAG pipelines.
• Call Logging - Captures and logs every query, retrieval, and generation step from LLM applications for later inspection.
• Experiment Progress Monitors - Reports live status and progress of running experiments and sends notifications on failures or stalls.
• Python Expression Filters - Filters and groups logged data using Python expressions for flexible ad-hoc analysis and aggregation.
• Experiment Artifact Storages - Saves images, audio, and text artifacts as files on disk, linked to experiment runs for later retrieval.
• Artifact Versioning - Stores and retrieves large binary artifacts produced during training, linking them to specific runs.
• Run Filtering Queries - Implements a structured query syntax to filter and retrieve runs by parameters, metrics, or tags.
• Experiment Run Grouping - Groups related training runs under named experiments with tags and parameters for structured navigation.
• Programmatic Run Filters - Retrieves and filters logged experiment data through a structured query language for automated analysis.
• Remote Run Controllers - Starts, stops, and monitors training runs on a remote server with SSL support for secure communication.
• Reproducible Run Configurations - Captures full run configurations and environments to enable exact re-execution of training runs.
• Context Manager Wrappers - Provides a Python library with context managers that wrap run lifecycle for easy integration into training scripts.
• Data Type Explorers - Loads specialized UI explorers for metrics, images, audio, and text as plugins for visual analysis.
• LLM Performance Monitoring - Records latency, token usage, and error rates for each LLM invocation to surface bottlenecks and regressions.
• System Resource Tracking - Records CPU, memory, and other resource metrics alongside experiment runs to correlate performance with training behavior.
• Log Artifact Explorers - Provides a unified interface to browse, filter, and deep-dive into any logged artifact type across all sessions.
• Experiment Management - Records and compares thousands of machine learning training experiments.
• General Machine Learning - Open-source AI metadata and experiment tracker.
• Machine Learning - Tracks metadata for machine learning experiments and training runs.
• Experiment and Data Management - Tool for recording and comparing AI experiment metrics.
• Experiment Tracking - Tracks and visualizes machine learning experiments.
• MLOps and Workflows - Tool for recording and comparing ML training experiments.

Star history