What has this model been trained on?
#3
by
LouisHernandez
- opened
Hello,
I was trying to reproduce the experiments for this model, and I can't get results as good as the pre-trained one. I found the hyperparameters in the tensorboard, but I keep getting CUDA Out Of Memory on an instance with 4 A10G GPUs. So, my question is the one in the title : What has this model been trained on?
Thank you in advance.
LouisHernandez
changed discussion title from
What has this model been trained on ?
to What has this model been trained on?
Hi @LouisHernandez ,
It has been trained on RTX 8000 with 48GB of memory. You could reduce the batch size to better suit the available memory on your GPUs.