Can this model be trained on bert large?

[Umang9427]

No description provided.

[ranjeetds]

Yes. While running run_ner.py script change --bert_model argument to --bert_model=bert-large-cased/uncased

[Umang9427]

Yeah, but it will require higher specs of CUDA in machine, even google colab is not enough.

[ranjeetds]

You can always run it with a smaller batch size. You need to set --train_batch_size argument while running the script. For example, In my custom data case, some of the sentences are very long and hence I have to set max_seq_length to 512. I am using RTX 2080Ti GPU and with a batch size of 2 I can finish one epoch in ~4 hours (For ~60K training samples) for large bert model

[Swty13]

@ranjeetds

Hi ,
I am training my custom NER model using BERT with 1 lakh dataset.Could you plz help me what hardware configuration is required to train this.

( I have VM having 32GB and 64GB RAM would it be possible to train on this or any GPU configuration is must to train this.)