Hi, I’d like to know if the format of the dataset and the process to fine tune a model is the same if we are using a model able to keep a conversation (chatgpt style) and one that just answers one shot questions.
I saw many tutorials of developers fine tuning instruction models like T5 to act like a chatbot. If I understood it well, these models will never be able to keep a conversation. They will only answer questions independently. Am I right?
If I want to fine tune a conversational model which one do you advise to play with and is there a tutorial somewhere? I’ve been playing with fine tuning blenderbot small but I couldn’t find any documentation on how the dataset should be or an example of people who did it to check the results.
P.S: I want to fine tune a conversational model to be able to answer questions of a specific domain. Let’s say medicine, for example. I was thinking about using this dataset: medalpaca/medical_meadow_medical_flashcards · Datasets at Hugging Face
Thanks