Resume_from_checkpoint not work

Trainer
1

I have download a ckpt file from github,and I tried to load it to my model,it seems not work.Even I give a fake filename it can still run.So I suppose it is not working at all
My version of pytorch-lighting is 1.2.9,and I use resume_from_checkpoint in trainer.

2

image
image982×225 14 KB

3

Hi

1.2.9 is a bit too old for me to give good suggestions, but if I remember correctly the resume_from_checkpoint argument only used to load when you called trainer.fit(). Maybe you didn’t call fit(), maybe you are running .test().

You can alternatively load just the model weights using

model = YourModelClass.load_from_checkpoint("path/to/file")

In the recent versions of Lightning this has completely changed. Each Trainer method can load a checkpoint file through the ckpt_path argument.

Documentation

5 
# default used by the Trainer
trainer = Trainer(resume_from_checkpoint=None)

# resume from a specific checkpoint
trainer = Trainer(resume_from_checkpoint="some/path/to/my_checkpoint.ckpt")

I try to use this, but it is not working.And the version is because of the github project
saic-mdal/lama: :llama: LaMa Image Inpainting, Resolution-robust Large Mask Inpainting with Fourier Convolutions, WACV 2022 (github.com)

6

Actually,I have found the problem that my path was wrong :rofl:.However,I have noticed that this cannot handle the situtation that the model has been changed.It can’t force the load,while torch.load has the flag “strict”,so I can just set the flag to True