List your company

Automated Theorem Proving on Metamath (val)

56.5Performance700m policy+value a = 32
19.746429.288238.8348.3718Sep 7, 2020

Evaluation results

MethodLinks
700m policy+value a = 32Parameters=700m, Test-time compute=Increased, Alpha=322020.0956.59.2———
700m policy+valueParameters=700m, Optimization=Iterated learned value function2020.0947.214.6———
700m WebMathParameters=700m, Pre-training=WebMath2020.0942.5610.9———
700mParameters=700m2020.0931.582.5———
160mParameters=160m, Architecture=Transformers2020.0928.967.8———
MetaGen-ILDescription=Baseline and state of the art.2020.0921.16————
Evaristetraining=online training2022.05——82.68181.2
Supervisedtraining=supervised2022.05———6165.4