Mathematics on GSM8K (held-out)
81.2AccuracyFixed Seed Image
Evaluation Results
| Method | Links | |
|---|---|---|
| Fixed Seed ImageModel=Qwen3.5-2B2026.06 | 81.2 | |
| Optimized Image (ART)Model=Qwen3.5-2B2026.06 | 81.2 | |
| Random ImageModel=Qwen3.5-2B2026.06 | 80.36 | |
| LoRA (DAPO)Model=Qwen3.5-2B2026.06 | 77.33 | |
| BaselineModel=Qwen3.5-2B2026.06 | 76.65 | |
| Random StringModel=Qwen3.5-2B2026.06 | 71.49 | |
| Optimized Image (ART)Model=Qwen3.5-0.8B2026.06 | 58.53 | |
| Fixed Seed ImageModel=Qwen3.5-0.8B2026.06 | 56.33 | |
| Random ImageModel=Qwen3.5-0.8B2026.06 | 54.59 | |
| LoRA (DAPO)Model=Qwen3.5-0.8B2026.06 | 49.51 | |
| BaselineModel=Qwen3.5-0.8B2026.06 | 39.65 | |
| Random StringModel=Qwen3.5-0.8B2026.06 | 25.25 |