KnowGPT: Knowledge Graph based Prompting for Large Language Models

Authors: Qinggang Zhang, Junnan Dong, Hao Chen, Daochen Zha, Zailiang Yu, Xiao Huang

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on three benchmark datasets demonstrate that Know GPT significantly outperforms all competitors including the state-of-the-art Graph RAG models.
Researcher Affiliation Academia The Hong Kong Polytechnic University, Rice University, Zhejiang Lab
Pseudocode No The paper describes its methods in prose and uses mathematical equations but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes We have provided the code for the framework, accessible via this anonymous URL: https://anonymous.4open.science/status/Know GPT-DD64.
Open Datasets Yes Datasets. We evaluate Know GPT on three QA datasets spanning two fields: Commonsense QA [66] and Open Book QA [52] serve as benchmarks for commonsense reasoning, while Med QA-USMLE [34] acts as a domain-specific QA benchmark.
Dataset Splits Yes The statistics of these three datasets can be found in Table 5 in the Appendix. Table 5: The statistical information of three datasets. Dataset Question Choices Train Dev Test
Hardware Specification Yes All models are implemented in Pytorch and trained on an RTX 3090 with 24 RAM.
Software Dependencies No The paper mentions 'All models are implemented in Pytorch' but does not specify a version number for Pytorch or any other software dependencies with their versions.
Experiment Setup No The paper states it uses policy gradient and gradient clipping, and mentions using a random seed for runs, but it does not provide specific hyperparameter values (e.g., learning rate, batch size, number of epochs) for its models or for training.