You can run this colab notebook on a free google colab - T4 GPU. I've reduced the number of training data so that anyone should be able to successfully train and inferencing yourself within like 5 min time.
The model will give much better accuracy if you increase the number of dataset. It might take longer time but you should definitely give a try. Below is the code where you can increase the number of training dataset.
Just update the number 1500 to any number if you want to traing for longer period of time to get more accurate result.
In similar way, you can also increase validation dataset below. You can find this line of code block is in step1 of colab notedbook.
raw_train_dataset, rt_to_skip = random_split(train_dataset, [1500,len(train_dataset)-1500]) raw_validation_dataset, vt_to_skip = random_split(validation_dataset, [50,len(validation_dataset)-50])