Creative Writing on CreativeWriting v3
73.95LLM JudgeSPARD
Evaluation results
| Method | Links | |
|---|---|---|
| SPARDBackbone=Qwen3-8B2026.04 | 73.95 | |
| + DPOBackbone=Qwen3-8B2026.04 | 73.15 | |
| + GRPOrmBackbone=Qwen3-8B2026.04 | 72.75 | |
| + GRPOimpBackbone=Qwen3-8B2026.04 | 72.45 | |
| + GRPOavgBackbone=Qwen3-8B2026.04 | 72.16 | |
| GOLFModel Backbone=Qwen-3-8B2026.03 | 70.78 | |
| BaseBackbone=Qwen3-8B2026.04 | 69.9 | |
| Direct-LikertModel Backbone=Qwen-3-8B2026.03 | 69.56 | |
| Rubric-as-RewardModel Backbone=Qwen-3-8B2026.03 | 69.21 | |
| Rubric-as-RewardModel Backbone=Llama-3.1-8B-Instruct2026.03 | 68.12 | |
| Pairwise-GRPOModel Backbone=Qwen-3-8B2026.03 | 68.08 | |
| Critique-GRPOModel Backbone=Qwen-3-8B2026.03 | 67.3 | |
| GOLFModel Backbone=Llama-3.1-8B-Instruct2026.03 | 66.21 | |
| Critique-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 65.76 | |
| Direct-LikertModel Backbone=Llama-3.1-8B-Instruct2026.03 | 64.1 | |
| Qwen-3-8BModel Backbone=Qwen-3-8B2026.03 | 63.27 | |
| Pairwise-GRPOModel Backbone=Llama-3.1-8B-Instruct2026.03 | 62.95 | |
| + SFTBackbone=Qwen3-8B2026.04 | 57.5 | |
| Llama-3.1-8B-InstructModel Backbone=Llama-3.1-8B-Instruct2026.03 | 53.96 | |
| SPARDBackbone=Qwen2.5-7B-Instruct2026.04 | 52.49 | |
| + GRPOavgBackbone=Qwen2.5-7B-Instruct2026.04 | 50.89 | |
| + DPOBackbone=Qwen2.5-7B-Instruct2026.04 | 50.49 | |
| + GRPOrmBackbone=Qwen2.5-7B-Instruct2026.04 | 50.17 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.04 | 48.09 | |
| + GRPOimpBackbone=Qwen2.5-7B-Instruct2026.04 | 45.88 | |
| + SFTBackbone=Qwen2.5-7B-Instruct2026.04 | 41.85 |