DeepChecks

Validation suites for machine learning models and datasets are the core of this Python package: wrap training and test data in its dataset objects, pick a suite or single checks, and get a report flagging conflicting labels, duplicates, leakage between splits, feature and prediction drift, weak data segments and performance gaps. Separate modules cover tabular models, NLP classification and token tasks, and computer vision on PyTorch or TensorFlow, with vision still in beta. Checks accept conditions that turn results into pass, warn or fail, so a suite can gate a CI pipeline; output goes to an interactive HTML file, a Jupyter view, or JSON. Install is a pip command with optional nlp and vision extras. A companion monitoring server self-hosts with Docker for one model per deployment; more than that, plus premium features, needs a commercial license. The library is AGPL 3.0, which matters if you modify it inside a network service. Releases have slowed, the last tagged version dating to December 2024 with occasional fixes since, so it suits teams wanting rigorous pre-deployment tests for classic ML more than LLM evaluation.