How does Query files work actually?

[KansaiTraining]

I have been trying to use PrivateGPT for a while and quite frankly I still don't get it. I asked about the three execution modes and got no answer sadly. I am specially puzzled by the "Query files" option. So I would appreciate some guidance on this more detailed problem.

I have a csv file with two columns. The first one holds source file names , the second one descriptions in natural language. I have ingested this document and would like to ask about it and get response. Can someone give me examples of appropriate queries?

---

Right now I ask for example "What can you tell me about the document" The answer

Based on just the provided context, which consists of ten separate observations describing various features and conditions in relation to an image or photograph, it is difficult to determine much about the larger document from which these observations have been extracted. The text suggests that the scene depicted in the image is a nighttime road scene with buildings, street signs, tall structures, parked vehicles, traffic lights, and an overcast sky. There are also mentions of "processed_images" and a watermark indicating copyright. However, without more context it's impossible to say for certain what the document as a whole is about or why these observations were made in the first place. It could be part of a larger report or analysis, or simply notes taken during an observation exercise. Without further information, any conclusions drawn from this limited context would be speculative at best.

Which is highly wrong. The system seems to think that it is about one image, when in fact there are more than 200 descriptions of images

Ideally I would like to ask something

"How many of the images contain street lights?" and get a good response but now that seems quite far from being accomplished

Also in the GUI what is "Selected for query or deletion"? Do I have to select a document to query about it? Ingestion is not enough?