KL Divergence Evaluation on SD 3.5
1.16KL Divergence (x10^-3)Flow-DPPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Flow-DPPOSampling schedule=Flow-SDE, Evaluation setup=Single2026.06 | 1.16 | |
| Flow-DPPO + CPSSampling schedule=CPS, Evaluation setup=Single2026.06 | 1.6 | |
| GRPO-GuardSampling schedule=Flow-SDE, Evaluation setup=Single2026.06 | 2.05 | |
| Flow-GRPOSampling schedule=Flow-SDE, Evaluation setup=Single2026.06 | 2.34 | |
| Flow-CPSSampling schedule=CPS, Evaluation setup=Single2026.06 | 2.41 | |
| Flow-DPPOSampling schedule=Flow-SDE, Evaluation setup=Multi2026.06 | 2.49 | |
| Flow-DPPO + CPSSampling schedule=CPS, Evaluation setup=Multi2026.06 | 2.52 | |
| Flow-CPSSampling schedule=CPS, Evaluation setup=Multi2026.06 | 3.18 | |
| GRPO-GuardSampling schedule=Flow-SDE, Evaluation setup=Multi2026.06 | 3.33 | |
| Flow-GRPOSampling schedule=Flow-SDE, Evaluation setup=Multi2026.06 | 3.81 |