I used gpt-2-keyword-generation to take my dataset and tokenize it. In the end, the file was about 700MB. When I try to train with any model size, the colab notebook runs out of memory. I know my dataset is pretty big, but is there anything I can do to get away with it?
Not on free tier, as far as I know. They used to give extra memory if your instance crashed but they don't now.
You could try and use the TPU-edited version.
I used gpt-2-keyword-generation to take my dataset and tokenize it. In the end, the file was about 700MB. When I try to train with any model size, the colab notebook runs out of memory. I know my dataset is pretty big, but is there anything I can do to get away with it?