I am making a langchain app which reads multiple pdfs and then users can chat with those pdfs. I am using OpenAIEmbeddings and ChatOpenAI as conversation chain. I have not bought openai subscription and using free credits, but I am facing usage limits from openai as i try to upload a mere 1.63 MB pdf file for embedding and then conversation. I want to know is the api limit really like this or am i doing something wrong as I think these usage limits are just not really practicle for real world applications. Also when I uploaded a very small file(0.7KB), it worked, but the responses were not of quality of chatgpt. why it is like that when i am using openai models?
Free accounts have fairly small usage limits. 48 hours after adding a credit card you’ll be increased to normal limits. You can see your rate-limits here.
Also when I uploaded a very small file(0.7KB), it worked, but the responses were not of quality of chatgpt.
This thing is more of an art than a science. Things that can affect the quality: how well the PDF is converted to text (some PDFs are designed poorly), the size of your chunks for embeddings, the prompt you/Langchain use, the temperature setting, and many others. It takes a lot of tweaking and experimenting and examining the quality to get things dialed in correctly for your specific use case.
2 Likes