I want to load the CKPT file for validation in trainer.validate, but it cannot work due to parameter mismatch. Where should I modify it to disable keyword matching.
trainer.validate(module,
datasetmodule,
ckpt_path=args[‘ckpt_path’]
)
I have defined the follow definition in my pl.LightningModule : if ckpt_path is not None:
print(“Load pretrained weights of backbone.”)
ckpt_dict = torch.load(ckpt_path)
self.load_state_dict(ckpt_dict[‘state_dict’], strict=False) But it seems useless.
And then I add the ckpt path to call the pl.LightningModule:
module = module.Network(network,
criterion,
cfg[‘train’][‘learning_rate’],
cfg[‘train’][‘weight_decay’],
val_step_settings=cfg[‘val’][‘step_settings’],
ckpt_path=args[‘ckpt_path’]
)
But it seems useless.
If you have some good edvices, please inform me. Thank you !