Hi
I finetune xml-roberta-large according to this tutorial. I met a problem that during training colab CUDA is out of memory.
RuntimeError: CUDA out of memory. Tried to allocate 978.00 MiB (GPU 0; 14.76 GiB total capacity; 12.62 GiB already allocated; 919.75 MiB free; 12.83 GiB reserved in total by PyTorch
And it is given that batch_size = 1
I tried to do that on xml-roberta-base, training lasts longer but over all ends up with the same problem. I tried bert-base-uncase that is in tutorial and it’s okay. But my data is multillingual!
I want to understand is it true that this problem is just because of natural limits of Colab or it is my fault. Is it possible to finetune xml roberta large in Colab?
Thanks!