I am following the tutorial https://github.com/ncoop57/i-am-a-nerd/blob/master/_notebooks/2020-05-12-chatbot-part-1.ipynb on fine-tuning DialoGPT (GPT-2) with a new conversational dataset.
It was trained fine earlier, the perplexity was about 5, 6 and the resulting dialogue seemed normal. enter image description hereNow, I am not sure if I touched somewhere...when I plotted the training loss, I found it went down to below 1 (about 0.25) and perplexity was over 300?
