json oserror to load the tokenizer due to the current os 'with_pretrained' A more optimized model according to the dataset.