Training model save problem

[xcxhy]

Describe the feature

Hello, I use the ChatGPT/examples/train_dummpy.py to train GPT2, and the training is successful, but the trained model cannot be found. Can you write a document to explain the problems of saving the model and using the training model?

[terminator123]

i met the same problem too

[ht-zhou]

Thanks for your feedback, but I have to remind you that train_dummy.py using random input, and you may not find any improvement after training.
Anyway, we'll add some guidance and demo to help save and inference the model.

[HuggingLLM]

@ht-zhou I have read source code, maybe train_prompts.py also not save the checkpoint. please add more demo help us to run it.

[xcxhy]

@NLP-ZY Agree. I notice is that neither train_dumm.py nor train_prompts.py save the checkpoint.

[YeHaijia]

same

[lajiyuan]

same

[Syno8]

I mkdir first as follows and save the model correctly.
os.makedirs(os.path.dirname(args.save_path), exist_ok=True)

[A-biao96]

i met the same problem too

Hello, did you solve the problem?

[FrankLeeeee]

This issue shall be fixed in #2846 once it is merged.

[A-biao96]

i met the same problem too

Hello, did you solve the problem?
This issue shall be fixed in #2846 once it is merged.

Thanks, I got it.