Question Answering on GPQA (Accuracy, Speedup)
31.03AccuracyProphet
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ProphetBackbone=LLaDA Base2025.12 | 31.03 | 15.32 | |
| Exp-k=16 (7.5, 2.5)Scheduling Strategy=Exponential (SchED), Curvature (k)=16, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 30.8 | 4.47 | |
| Cosine (7.5, 0)Schedule Type=Cosine, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 30.58 | 15.96 | |
| Cosine (7.5, 2.5)Schedule Type=Cosine, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 30.58 | 15.93 | |
| Linear (7.5, 0)Schedule Type=Linear, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 30.58 | 16.08 | |
| Linear (7.5, 2.5)Schedule Type=Linear, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 30.58 | 16.01 | |
| Exp-k=2 (7.5, 0)Schedule Type=Exponential, k=2, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 30.58 | 16.32 | |
| Exp-k=16 (7.5, 2.5)Schedule Type=Exponential, k=16, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 30.58 | 17.16 | |
| BaselineSchedule=Standard2025.12 | 30.36 | 1 | |
| Exp-k=2 (7.5, 2.5)Schedule Type=Exponential, k=2, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 30.13 | 16.18 | |
| Exp-k=16 (7.5, 0)Schedule Type=Exponential, k=16, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 29.46 | 18.64 | |
| Exp-k=2 (7.5, 0)Scheduling Strategy=Exponential (SchED), Curvature (k)=2, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 29.02 | 2.85 | |
| ProphetSchedule=Prophet2025.12 | 28.79 | 16.45 | |
| Exp-k=16 (7.5, 0)Scheduling Strategy=Exponential (SchED), Curvature (k)=16, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 28.79 | 8.67 | |
| Exp-k=4 (7.5, 2.5)Scheduling Strategy=Exponential (SchED), Curvature (k)=4, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 28.79 | 2.86 | |
| Exp-k=8 (7.5, 2.5)Scheduling Strategy=Exponential (SchED), Curvature (k)=8, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 28.57 | 3.7 | |
| Linear (7.5, 0)Scheduling Strategy=Linear, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 28.57 | 2.35 | |
| Exp-k=8 (7.5, 0)Scheduling Strategy=Exponential (SchED), Curvature (k)=8, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 28.35 | 5.94 | |
| Linear (7.5, 2.5)Scheduling Strategy=Linear, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 28.12 | 1.98 | |
| Cosine (7.5, 0)Scheduling Strategy=Cosine, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 27.9 | 2.24 | |
| Exp-k=4 (7.5, 0)Scheduling Strategy=Exponential (SchED), Curvature (k)=4, Thresholds (tau_high, tau_low)=(7.5, 0), Backbone=LLaDA Base2025.12 | 27.46 | 4.09 | |
| Cosine (7.5, 2.5)Scheduling Strategy=Cosine, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 27.23 | 1.97 | |
| Exp-k=2 (7.5, 2.5)Scheduling Strategy=Exponential (SchED), Curvature (k)=2, Thresholds (tau_high, tau_low)=(7.5, 2.5), Backbone=LLaDA Base2025.12 | 26.34 | 2.24 | |
| BaselineBackbone=LLaDA Base2025.12 | 26.12 | 1 |