#Feed GPT-3 API with Data-sets
4 messages · Page 1 of 1 (latest)
maybe this wil help you https://github.com/openai/openai-cookbook/blob/main/examples/Question_answering_using_embeddings.ipynb
Thanks, I am aware it works by giving it 100 questions and answers. But can I feed it a huge handbook for example? Anyone know if this works?
you can use embedding. The way embedding works is to first create vectors for your knowledge base, then when you pose a question you first search your knowledgebase for the closest vectors to your question, then take the top matches and feed them as prompts to the model (like davinci). I'm personally not very happy with embedding results from my experiments. But I think a combination of embedding and fine-tuning will create better results.