`The pretrained large BERT model has 340M parameters and is 1.35GB. Even if it can fit on a consumer device (e.g. your phone), the time it takes for BERT to run

`The pretrained large BERT model has 340M parameters and is 1.35GB. Even if it can fit on a consumer device (e.g. your phone), the time it takes for BERT to run inference on a new sample makes it useless for many real world applications. ` This is true, but a fine-tuned Distilbert model will give close to that accuracy with a ~260MB file size.