generated: '2026-08-04'
method: searched
source: https://github.com/Paige-AI/paige-ml-sdk
name: paige.ml_sdk
invocation: python -m paige.ml_sdk
description: >-
  First-party command line interface shipped inside the Paige ML SDK for training
  and running inference with supervised computational-pathology models over tile-level
  embeddings (e.g. embeddings produced by the Virchow / Virchow2 foundation models).
  Built on the PyTorch Lightning CLI, so the full Lightning Trainer flag surface is
  exposed. There is no CLI for the commercial Paige Platform / UnifAI API.
package: packages/paige-packages.yml#paige-ml-sdk
install:
- method: source
  command: git clone https://github.com/Paige-AI/paige-ml-sdk && cd paige-ml-sdk && pip install -e .
help: python -m paige.ml_sdk --help
commands:
- name: fit
  description: Train a model against a dataset of tile-level embeddings.
  help: python -m paige.ml_sdk fit --help
flag_groups:
- name: trainer
  prefix: --trainer.
  description: >-
    PyTorch Lightning Trainer configuration — strategy (DDP/FSDP), accelerator, precision,
    epochs. Example documented in the README: --trainer.precision 16.
- name: logger
  prefix: --trainer.logger
  description: >-
    Logger selection. Defaults to Lightning's CSVLogger writing to ./lightning_logs;
    any Lightning built-in logger (WandB, TensorBoard) may be substituted.
- name: optimizer
  prefix: --optimizer
  description: Any torch.optim algorithm.
- name: lr_scheduler
  prefix: --lr_scheduler
  description: Learning-rate schedule selection.
- name: callbacks
  prefix: --trainer.callbacks
  description: >-
    Lightning callbacks. By default only a model-checkpoint callback is applied.
examples: https://github.com/Paige-AI/paige-ml-sdk/tree/main/examples
x-evidence:
  fetched: '2026-08-04'
  url: https://raw.githubusercontent.com/Paige-AI/paige-ml-sdk/main/README.md
  http_status: 200