MathSupra is a fine tune of Supra-50M-reasoning designed for math, it was fine tuned on 25k subset of mrfakename/basic-math-10m for 1 epoch.
MathSupra was fixed multiple times to be smarter.
Evaluation Metrics Summary
GSM8K 0-Shot Accuracy: 1.00% evaluated across standardized community slices tracked on the AxiomicLabs Open SLM Leaderboard.
