I'm new to ml, and trying to create a ml model fine tuning GPT2.
I get the dataset and preprocessed it (file_name
). But when I actually try to run below code, fine tuning GPT2, Colab always say 'Your session crashed after using all available RAM.'
gpt2.finetune(sess,
dataset=file_name,
model_name='124M',
steps=50,
restore_from='fresh',
run_name='run1',
print_every=10,
sample_every=10,
save_every=10,
batch_size=16
)
I'm already on the Colab pro version and RAM size is 25GB, and file size is only about 500MB. I tried lowering training steps and batch size but this error is keep happening.
Any idea how can I stop this behavior?