Integrating 3LC with PyTorch

3LC integrates closely with PyTorch as a machine-learning framework. PyTorch powers the torch-dependent parts of 3LC, such as collecting per-sample metrics from a model with tlc.metrics.collect.collect_metrics() or creating torch samplers using a sample weight column.

Note

As of 3lc >= 3.0, PyTorch is an optional dependency. The core surface — Tables, Runs, the Object Service, and URL adapters — works without torch installed, while the torch-dependent entry points raise an ImportError pointing at pip install 3lc[torch] when called without it. See PyTorch on the dependencies page for the full list of torch-dependent entry points and for installation guidance, including accelerator-specific wheels.