🕹 QA Experiment, various comparative experiments to make llm approach business experts step by step
Traditional industries transition to the AI era, the most valuable step is cleaning QA
This matter is divided into a three-step strategy
- First, for a specific tax field, (Yuran) proposes 3 real business problems, and then starts with the knowledge points and cases that need to be understood to solve the problems in this field, and forms 100 high-quality QAs . Then select other business problems in this field, and we do comparative experiments (Yawei) to cooperate
A. Ask OpenAI directly
B. Ask OpenAI after optimizing prompt words
C. Find relevant QA + FewShot and ask OpenAI
D. Put these 100 questions based on OpenAI FineTune and ask OpenAI
E. Based on these 100 QAs , fine-tune a Lora model based on ChatGLM and ask ChatGLM
F. Cooperate with cc to train all QA training based on tax basics and Yuran optimization from scratch on land Value Added Tax related tax models
-
Based on the above comparative experiment, we went to find more people in the consulting industry and told them the role of QA
-
Enterprise Alliance creates a convenient QA precipitation tool for people in the above industries to solve problems: how to form Q, how to modify A, how to classify QA