Knowledge file hierarchy issue: Knowledge in Workspace doesn’t support folders, folder name, file name & Model with knowledge can't find correct answer #3844

New Issue

2025-11-11T15:41:03-06:00

GiteaMirror commented

2025-11-11 15:41:03 -06:00

Originally created by @jerry0012000 on GitHub (Feb 14, 2025).

For two series of products, I created two knowledge bases, for example series A and series B. Then I uploaded product documents into knowledge bases, for example A1_basic_information, A1_datasheet, B1_basic_information, B1_datasheet.
I wish only use one knowledge base, called “product”, with series A and series B as subfolder, then subfolder of product A1, B1, then files for each product.

When I selected those two series(two knowledge bases), into my customized model, and starting chat with it, I noticed that the model can’t recognize the name of knowledge bases, even though I type exactly the name of product and ask for its price(which is inside the basic information), the model can’t find it from the correct file.

I tried “What is price of series A?” The model can’t navigate to correct knowledge base, it sometimes read B1_basic_information first, answered price of B1, and most of the time read B1_datasheet first, and didn’t reply the price(although it sometimes scanned A1_basic_information as third file).

Ideal answers are: “Product A1 belongs to series A, here is its price: [Price of A1].”

Similar phenomenon happened when I try price of series B or product B1, it replied no price or price of A1.

Ideal answer of a certain product should be from correct file, but I neither think the model know the file name, nor the folder(Knowledge Base) name.

Model I used are deepseek-r1:1.5b and deepseek-r1:7b, I didn’t change any parameters or settings.

Should I use other models, give up RAG solutions, or finetune deepseek models with those documents?

Can find correct answer from uploaded files.

Can't find correct answer, generate ransom answer, fail to answer from model with knowledge.

Originally created by @jerry0012000 on GitHub (Feb 14, 2025). For two series of products, I created two knowledge bases, for example series A and series B. Then I uploaded product documents into knowledge bases, for example A1_basic_information, A1_datasheet, B1_basic_information, B1_datasheet. I wish only use one knowledge base, called “product”, with series A and series B as subfolder, then subfolder of product A1, B1, then files for each product. When I selected those two series(two knowledge bases), into my customized model, and starting chat with it, I noticed that the model can’t recognize the name of knowledge bases, even though I type exactly the name of product and ask for its price(which is inside the basic information), the model can’t find it from the correct file. I tried “What is price of series A?” The model can’t navigate to correct knowledge base, it sometimes read B1_basic_information first, answered price of B1, and most of the time read B1_datasheet first, and didn’t reply the price(although it sometimes scanned A1_basic_information as third file). Ideal answers are: “Product A1 belongs to series A, here is its price: [Price of A1].” Similar phenomenon happened when I try price of series B or product B1, it replied no price or price of A1. Ideal answer of a certain product should be from correct file, but I neither think the model know the file name, nor the folder(Knowledge Base) name. Model I used are deepseek-r1:1.5b and deepseek-r1:7b, I didn’t change any parameters or settings. Should I use other models, give up RAG solutions, or finetune deepseek models with those documents? Can find correct answer from uploaded files. ![Image](https://github.com/user-attachments/assets/d92cef82-1f04-4444-a559-73d93a264aa9) Can't find correct answer, generate ransom answer, fail to answer from model with knowledge. ![Image](https://github.com/user-attachments/assets/dcc28d9a-3b45-4f39-b43c-6db879ad1bfc)

GiteaMirror closed this issue

2025-11-11 15:41:03 -06:00

GiteaMirror commented

2025-11-11 15:41:03 -06:00