ResearchBenchmarksMathematical Reasoning on GSM8K (test) (Accuracy, n, MSE1)Follow73AccuracyB266.65668.30369.9571.597Apr 20, 2026Evaluation ResultsMethodMethodLinksAccuracySample Size (n)MSE1B2Configuration=LM only,...Configuration=LM only, Decoding=greedy2026.04739630.6AConfiguration=consec-s...Configuration=consec-step, Decoding=greedy2026.04739630.006A1Configuration=STP only...Configuration=STP only, Decoding=greedy2026.0469.39140.006B1Configuration=frozen,...Configuration=frozen, Decoding=greedy2026.0468.89070.955CSTP Type=rand-tok, Dec...STP Type=rand-tok, Decoding=greedy2026.0466.98820.226