Code generation on LiveBench (Pass@K, GSS@K)
20.39Pass@10Qwen2.5-VL-72B-Ins
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-VL-72B-Ins2025.09 | 20.39 | — | 22.37 | — | |
| GRPOTN=16, AN=1, S/S=186.912025.09 | 11.92 | 6.3 | 14.12 | 25 | |
| GRPO-MATN=4, AN=4, S/S=93.452025.09 | 11.69 | 6.73 | 14.7 | 10 | |
| GRPOTN=4, AN=1, S/S=76.212025.09 | 11.56 | 6.59 | 13.7 | 6 | |
| GRPOTN=8, AN=1, S/S=104.832025.09 | 11.44 | 6.53 | 13.39 | 24 | |
| Qwen2.5-VL-7B-Ins2025.09 | 10.72 | — | 11.31 | — | |
| Qwen2.5-VL-3B-Ins2025.09 | 9.8 | — | 11.67 | — | |
| SFT2025.09 | 8.72 | — | 10.59 | — |