I’ve been trying to load a checkpoint from a trainer with the load_checkpoint() function, since it was saved by a save_checkpoint() function. The reason I am using these is because they are the only ones that allow me to save a model’s weights without doing any training. My project unfortunately requires that.
However, I am unable to load this checkpoint inside tune. Since it is not a tune-checkpoint, it is not a folder, thus it cannot be “restored”. The method to load it, is from the trainer by creating like :
trainer = PPOTrainer(config)
trainer.load_checkpoint(path)
However, this trainer cannot be loaded inside tune like this:
tune.run(trainer, other_configs...)
because it is not “string nor trainable”
Can anyone help me out with this?
Thank you in advance