List your company

Reasoning on Reasoning Dataset Regime III: Process-degraded

2.21Speedup (batch-1)loop rescue (+L)
0.15080.68541.221.7546Jun 1, 2026

Evaluation results

MethodLinks
loop rescue (+L)Setup=loop rescue, N=93572026.062.21—5.241.98
FP16 planning + loop rescue (+P + L)Setup=FP16 planning + loop rescue, N=93572026.061.81—8.81.46
2bitSetup=2-bit quantization, N=93572026.060.2367.89—0.22