Advances in few-shot learning: reproducing results in PyTorch