unitorch

Guidance for writing, reviewing, and updating unitorch INI configuration files for train, eval, infer, and FastAPI workflows. Use when creating or modifying examples/configs/*.ini, reasoning about Config interpolation and CLI overrides, composing preprocess_functions, choosing registered component names, or debugging unitorch CLI config behavior.

Use when an agent needs to invoke the `core/copilot/pkg_infos` unitorch copilot tool.

Use when an agent needs to discover and invoke registered UniTorch copilot tools through `unitorch-copilot-cli`.

Guidance for creating, reviewing, running, and debugging unitorch batch model inference workflows with unitorch-infer. Use when preparing inference INI configs, choosing registered models/processors/postprocess functions/writers, running inference from checkpoints, configuring vLLM inference, preserving raw input columns in outputs, or diagnosing empty, duplicated, or incorrect inference results.

Reference for using, reviewing, or modifying unitorch's @replace decorator in src/unitorch/utils/decorators.py. Use when overriding upstream classes, adding replacement classes under modules/replace, reasoning about process-global monkey patches, or debugging import-time replacement and subclass __bases__ rewriting.

Guidance for creating, reviewing, launching, calling, and debugging unitorch FastAPI services with unitorch-fastapi and CopilotClient. Use when preparing FastAPI INI configs, choosing registered core/fastapi services, configuring routers and pipeline sections, starting or stopping daemon/foreground servers, calling /start /status /generate endpoints, using autostart_services, or diagnosing health-check, port, route, service loading, and request-format issues.

Guidance for creating, reviewing, running, and debugging unitorch model training workflows with unitorch-train. Use when preparing training INI configs, choosing registered models/processors/losses/scores, launching single-GPU or distributed training, checking checkpoint loading, monitoring loss and validation metrics, or diagnosing unstable training.