THUDM

/

glm-4-9b-chat

like 621

Follow

Knowledge Engineering Group (KEG) & Data Mining at Tsinghua University 1,112

check our github and find openai_demo or composite demo

Hi, thanks for instruction. I followed the same approach as the OpenAI demo to generate the tool prompt, but the result of the function call was not as good as the result from the ZhipuAI API. Is the model on Hugging Face the same as the one used in your official API?

No, ZhipuAI API uses our commercial model, not the open source model, and the model size is different

Thanks for the explanation!

However, in the picture above, GLM-4-9B-Chat seems nearly as effective as GPT-4 in function calls. How significant is the difference between the open-source and commercial versions? Does this mean your commercial model is better than GPT-4 in function calls? Is there any performance metrics for the commercial model?

Thank you for keeping responding my questions, appreciate your time!

I conducted function call tests using my own test cases. The official API scored 92, while the open-source model only achieved a score of 65. Is this expected?