Language Generation on Goodreads Books (test)
61.38ROUGE-1P-GRPO
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| P-GRPOModel=Qwen3-8B2026.02 | 61.38 | 13.83 | 35.78 | 52.96 | |
| P-GRPOBackbone=Qwen3-8B2026.02 | 61.38 | 13.83 | 35.87 | 52.96 | |
| GRPOModel=Qwen3-8B2026.02 | 60.76 | 13.74 | 35.96 | 52.66 | |
| GRPOBackbone=Qwen3-8B2026.02 | 60.76 | 13.74 | 35.96 | 52.66 | |
| P-GRPOModel=Gemma-2B2026.02 | 55.34 | 11.81 | 32.04 | 54.15 | |
| P-GRPOBackbone=Gemma-2B2026.02 | 55.34 | 11.81 | 32.04 | 54.15 | |
| GRPOModel=Gemma-2B2026.02 | 55.26 | 10.6 | 32.55 | 53.28 | |
| GRPOBackbone=Gemma-2B2026.02 | 55.26 | 10.6 | 32.55 | 53.28 | |
| GDPOModel=Qwen3-8B2026.02 | 49.87 | 7.81 | 30.03 | 45.2 | |
| GDPOBackbone=Qwen3-8B2026.02 | 49.87 | 7.81 | 30.03 | 45.2 | |
| GDPOModel=Gemma-2B2026.02 | 40.08 | 6 | 22.54 | 52.88 | |
| GDPOBackbone=Gemma-2B2026.02 | 40.08 | 6 | 22.54 | 52.88 |