List your company

Mathematical Reasoning on GSM8K (Acc, TNFT, Tokens, Delay)

91.85AccuracyBase
88.7389.5490.3591.16May 12, 2026

Evaluation results

MethodLinks
BaseBackbone=Qwen3-4B2026.0591.85117.031,340.4941.94
VanillaBackbone=Qwen3-1.7B2026.0590.693.3660.6414.93
BaseBackbone=Qwen3-1.7B2026.0590.37117.031,156.127.14
OursBackbone=Qwen3-1.7B2026.0589.510437.111.29
VanillaBackbone=Qwen3-4B2026.0589.3693.3649.2820.17
OursBackbone=Qwen3-4B2026.0588.850421.4714.53