Resuming training

[infamix]

If I resume training from a ckpt the dataset will continue right where it left off, right?

[erfanzar]

No it won't start where it left off from dataset but if the data is saved by easydel savers itself optimizers will resume on previous ckpt and load from last step

[infamix]

Is this all that's needed to resume training?

trainer = CausalLMTrainer(
    train_args,
    dataset_train,
    ckpt_path="/kaggle/input/easydel-sq-6000/EasyDeL/lib/python/easydel_ckpt/EasyDelLLama2/EasyDelLLama2-1.5074825286865234-6000"
)

Because it didn't happen and I'm a little confused.

[erfanzar]

Yes that all you need and then you call .train function to train model

[infamix]

I understand but it didn't work as intended, instead the training started from 0.

[erfanzar]

Yes it will start from step 0 but resume the previous process

[infamix]

Negative, the loss and the optimizer graphs (etc.) looked exactly like you'd start from 0 and not like you'd resume from a checkpoint.