validation and checkpoints

[flyingflying]

When I use every_n_train_steps in ModelCheckpoint callback and val_check_interval in Trainer, I found that trainer would save checkpoints before validation.

What should I do to make sure that trainer will validate first, and then save checkpoints?

My codes are shown below:

model_checkpoint = pl.callbacks.ModelCheckpoint(
    monitor="02_eval_spearmanr", mode="max", dirpath=output_dir, save_top_k=2, every_n_train_steps=100,
)

trainer = pl.Trainer(
    max_epochs=1, accelerator="cuda", devices=[0, ], default_root_dir=output_dir,
    callbacks=[model_checkpoint, ], move_metrics_to_cpu=True, val_check_interval=100,
)