Vision-Language Modeling on DynaM
52.7AccuracyZPPO
Evaluation Results
| Method | Links | |
|---|---|---|
| ZPPOModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.06 | 52.7 | |
| GRPOModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.06 | 45.9 | |
| GRPOModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.06 | 45.6 | |
| On-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.06 | 44.9 | |
| On-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.06 | 44.5 | |
| Off-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=true2026.06 | 43.7 | |
| Off-DistillModel Scale=2B, Prompt Replay Buffer Augmentation=false2026.06 | 43.2 | |
| Qwen3.5Model Scale=2B, Prompt Replay Buffer Augmentation=false2026.06 | 41.1 | |
| ZPPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.06 | 31.1 | |
| GRPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.06 | 23.6 | |
| GRPOModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.06 | 21.3 | |
| On-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.06 | 21.1 | |
| Off-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=true2026.06 | 20.1 | |
| On-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.06 | 20 | |
| Off-DistillModel Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.06 | 19.4 | |
| Qwen3.5Model Scale=0.8B, Prompt Replay Buffer Augmentation=false2026.06 | 17.8 |