WizardCoder 34B surpasses GPT4 on HumanEval (opens in new tab)

(twitter.com)

1 pointsbytefactory2y ago1 comments

1 comments

1 comments · 1 top-level

WizardLM claims to have beaten GPT4 on HumanEval with 73.2% @ Pass1 to GPT4s 67% (in the original GPT4 report).

Big, if true. Their own testing puts present GPT4 at 82%, but 73.2% still seems very competitive.

j / k navigate · click thread line to collapse