Resuming training

[dimagalat]

Hello,

I'm trying to implement the following scenarios:

• load a learner to continue training on the same dataset
• load a learner to train on a different dataset

Unfortunately, I get the following error when trying to load a learner:
runtimeerror: pytorchstreamreader failed reading zip archive: failed finding central directory
I've read that load_learner, "requires all your custom code be in the exact same place as when exporting". Could someone clarify what that means in practice? Re-loading all the same objects in the environment?

Can anyone point out an example where a learner is used on a different dataset?

Thanks in advance, I really appreciate your help

[dimagalat]

learn.hf_model.save_pretrained("model")
hf_tokenizer.save_pretrained("model")
solves it

[ohmeow]

Ok cool ... yah that makes sense for transformers. I'll try to find some time to add this to blurr so a single method gets the job done. Thanks

…

On Sun, Jul 10, 2022 at 2:48 PM Dima ***@***.***> wrote: Closed #80 <#80> as completed. — Reply to this email directly, view it on GitHub <#80 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAADNMENKN7F4TQR4YD4EODVTNAJ3ANCNFSM53CIIEPA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>