I wanted to see what you could do to train GPT on custom data. Fine tuning was a dud but I was amazed at what composition tools like LangChain were able to accomplish even on a small dataset.
Great post. Super-helpful understanding the context as well as actual code that illustrates the mechanism.