pytorch-lightning

c76cec6b - Fix `ModelCheckpoint` with manual optimization and `every_n_train_steps` (#21239)

Fix `ModelCheckpoint` with manual optimization and `every_n_train_steps` (#21239) * Fix ModelCheckpoint with manual optimization and every_n_train_steps - Ensure checkpoints reflect the model state before optimization when using manual optimization - Add warning when pre-optimization state isn't saved - Update documentation to clarify the behavior with manual optimization Fixes #20947 * cleanup log after test * fix future warning with torch.load