Error when calling trainer.test(ckpt_path="best"): ModelCheckpoint is not configured to save the best model

[xiaoqingli0608]

I am trying to run the pretraining pipeline and encountered an issue when calling trainer.test(ckpt_path="best"). The training itself finishes successfully, but testing fails with a checkpoint-related error.
Training runs normally for all epochs. However, when reaching the testing stage, the following error occurs:“ValueError: .test(ckpt_path="best") is set but ModelCheckpoint is not configured to save the best model.”
no “best” checkpoint is saved, which causes .test(ckpt_path="best") to fail.

[Boltzmachine]

Pretraining is not designed for evaluating at the end of the training. You should load the saved checkpoint manually and finetune it on downstream tasks with what was specified in https://github.com/Graph-and-Geometric-Learning/MindLLM?tab=readme-ov-file#finetuning.