Reaction
SpartacusML aims for a near-100% GSM8K Platinum test score
Its creator believes earlier prototypes suggest four requirements are achievable, though a “10M” constraint may require compromises.
TLDR
A developer says a GSM8K solver they began working on for fun is becoming a project called SpartacusML. Their original goal was a 100% score on the Platinum test set. Based on earlier prototypes, they think they may be able to meet four requirements, including a roughly 100% score, but say the “10M” constraint will likely take clever compromises.
Combined views
—
1 Source, first seen ago
