Open Ahmetay882 opened 3 weeks ago
Have you gone through our guide already about debugging answer quality?
That should help you pinpoint whether the issue is the retrieval, the LLM step, or the data formatting from the ingestion step.
As for #2), I suggest saying that in the system prompt. You can tell it that it's also allowed to use information that it already knows, if it can't find the answer in the sources.
Hello Pamela,
Thank you for answering my question.
I have solved my second issue, and I appreciate your help.
Regarding my first issue, I have looked at the steps, but I think the problem is caused by not being able to read the Excel file I uploaded. This issue does not occur when I upload a PDF document. Do you have any ideas about this?
@Ahmetay882 in my experience llms don't work as well with tabular data. I've had to write scripts to turn excel files into pdfs with text, and it works much better after being converted to text. unfortunately.
@bp3000bp Hello, can you give me an example of how you did it?
This issue is for a: (mark with an
x
)I want to create a financial ChatGPT assistant that can help my clients with finance-related topics. I am using this repository, but I have encountered some issues. My problems are:
1.) Even though I have uploaded my Excel file to this repository, when I ask questions related to my data, I receive an error saying that the information cannot be found in the provided file, even though it exists in the Excel table. Could you help me understand why this is happening?
2.) I want the assistant to prioritize using the data from my Excel table for finance-related questions. If the information is not available in my table, I would like it to fall back on GPT-4’s own data, but only for financial topics. How can I set this up?
I would greatly appreciate any help or guidance on these issues. Below, I am sharing a sample from a part of my Excel table.