Wow! 😱🚀
WizardCoder 34B outperformed GPT4, brad and claude on HumanEval. It is fine tuned version of Code Llama ( introduced yesterday) !

The 67% by GPT4 is the official report of Open.ai and the 72.5%(GPT3.5) and 82% (GPT4) are by the wizardlm team.

#crazyworld!

All the codes and analysis are opens source here

#چه_خبرتونه! #اونوقت_خبر_بالا😔