Code Generation on LiveCodeBench (avg@3 accuracy)
56.12Avg@3 AccuracyQwen3-8B-Base + PASC-GRPO
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen3-8B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 56.12 | |
| Qwen3-8B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 53.93 | |
| Qwen3-4B-Base + PASC-GRPOmax_completion_tokens=8K2026.01 | 46.79 | |
| Qwen3-4B-Base + GRP-SFTmax_completion_tokens=8K2026.01 | 42.93 | |
| Qwen3-8B-Basemax_completion_tokens=8K2026.01 | 29.76 | |
| Gemma-3-27B-ITmax_completion_tokens=8K2026.01 | 26.9 | |
| Gemma-3-12B-ITmax_completion_tokens=8K2026.01 | 25.7 | |
| Qwen3-4B-Basemax_completion_tokens=8K2026.01 | 21.14 | |
| LLaMA-3.1-8B-Instructmax_completion_tokens=8K2026.01 | 10.8 |