List your company

Mathematical Reasoning on GSM8K (EMA, SIP, TAG, LAMP, BiSR Metrics)

46.9EMANon-split
17.57225.18632.840.414Jun 15, 2026

Evaluation results

MethodLinks
Non-splitBackbone=Qwen3-1.7B2026.0646.97678.274.38474.483.942.434.7
DP-forwardBackbone=Qwen3-1.7B2026.0646.75756.5748373.683.539.835.9
DP-SGDBackbone=Qwen3-1.7B2026.0646.173.877.674.283.974.383.741.834.4
Naive MixBackbone=Qwen3-1.7B2026.0646.146.142.648.140.955.148.329.924.4
MIXGUARDBackbone=Qwen3-1.7B2026.0645.54.22.430.334.927.230.510.66.6
Non-splitBackbone=Llama 3.2-1B2026.0625.469.87074.682.175.782.534.429.4
MIXGUARDBackbone=Llama 3.2-1B2026.0623.63.72.630.334.627.330.517.414
DP-forwardBackbone=Llama 3.2-1B2026.0623.258.556.674.882.275.682.635.729
Naive MixBackbone=Llama 3.2-1B2026.0622.538.735.857.856.452.14527.721.1
DP-SGDBackbone=Llama 3.2-1B2026.0618.758.556.748.559.34558.321.221