I think the key lies in the components of the knowledge graph, and there is not a good external knowledge graph

zhao-zilong / ssc-cot

Git for "Stepwise Self-Consistent Mathematical Reasoning with Large Language Models"

MIT License

10 stars 0 forks source link

I think the key lies in the components of the knowledge graph, and there is not a good external knowledge graph #2

Open hxypqr opened 3 months ago

hxypqr commented 3 months ago

Thank you for sharing this enhanced mathematical reasoning approach for LLM. In addition, mathematics is like playing chess with the devil. Do you have a system that fine-tunes the entire SSC-CoT model with the signal returned by GPT4? Will this improve the performance of the model? I did not see any discussion on this in the paper. Thank you in advance for your response.

zhao-zilong commented 3 months ago

Hi @hxypqr we did not fine-tune any model. In ssc-cot, we propose a way to better prompt LLM with more relative knowledge.