Some questions about the training process

[Maydaytyh]

Thank you for your great work!
I couldn't find code in the repository that processes the retrieved content; the training data under LLMs/data_id appears to be identical to that used in ChatKBQA. Could you please point me to where the retrieved knowledge (entities, relations, subgraphs) is actually handled or transformed before being fed into the model? Thank you!

[quqxui]

We do not include the code for handling retrieved knowledge. We directly use the implementations from:
https://github.com/HXX97/GMT-KBQA and https://github.com/awslabs/decode-answer-logical-form

We rely on their code to transform the retrieved knowledge (entities, relations, and subgraphs). You can download the necessary resources from our Google Drive:
https://drive.google.com/drive/folders/1uOcpPoBcFeL2JWE6-Wpj6kuR-7Vdgiyz?usp=sharing

Additionally, the training data under LLMs/data_id is different from ChatKBQA because our method needs to add a unique data ID to each sample.