Reward modeling on EVAL_INSTRUCT 2 steps
1.65Step Completion RateR2VLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| R2VLM2026.03 | 1.65 | 70 | |
| Qwen2.5-VL-InstructModel Size=7B2026.03 | 1.55 | 65 | |
| Pretrained SPRINT2026.03 | 1.35 | 55 | |
| Step-Completion Based Reward2026.03 | 1.3 | 55 |
Loading the SOTA2 catalog…