search

KevinBian107 / RPLH

Developing frameworks for multi-agent collaborations using Role-playing Leader-hellucination System
https://kbian.org/RPLH/
2 stars 0 forks source link

Better synthetic check #13

Open KevinBian107 opened 4 hours ago

KevinBian107 commented 4 hours ago

Make synthetic check better and re better for getting json format answer

KevinBian107 commented 3 hours ago

Current progress is that there was a error in the Json synthetic check where we didn't specify the format of the output json file, so the checking process caused hellucinating issues of LLM, after specifying format and fixing the flow of info for judge and construction of messages, the pipeline is runable, now need to figure out how to converge, need visualization and tunning