How to Save Notebook (With GPU state etc)

Part 1 (2018)
1

  1. How can I save the notebook to load it later? If I hibernate the notebook, it doesn’t seem to work after I reboot.

  2. Alternatively, how can I save the weights to reload them later? If I save the weights like in lesson 1 (such as : " learn.save('224_lastlayer') "), I still have to retrain to load new ones (requiring rer-running : " learn = ConvLearner.pretrained(arch, data, precompute=True) "). Isn’t there a way to just dump the whole thing so it can be created based on a saved file maybe with h5py or something?

2 Likes
2

fastai provides a very nice api for this :slight_smile:

learn.save('file_name')
learn.load('file_name')
1 Like
3

yes, but how can you create the Object “learn” without training again?

4

You initialize the model how you initialized it initially (define layers, use pretrained arch, etc) and then call load on it.

1 Like
5

Is this the best way then?

6

If this was how you created the learner that you saved, than I think that would be the way to go.

1 Like
7

so by doing that, we can skip doing “learn.fit” again?

8

yes, your model should be able to make predictions like at the moment when you called save on it, the weights should be restored

2 Likes
9

Hello, I wanted to know about loading the saved model in the enhance notebook.
m = to_gpu(SrResnet(64, scale))
m = nn.DataParallel(m, [0])
learn = Learner(md, SingleModel(m), opt_fn=optim.Adam)
learn.load(“sr1-a”)

Is this the correct way?
Getting the following error!
KeyError: ‘unexpected key “features.0.0.weight” in state_dict’