Factual Knowledge Evaluation on Wikidata knowledge infusion
64.69AccuracyPretrainRL
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| PretrainRLBase Model=Llama3-8B, Training Method=PretrainRL, Decoding Strategy=Beam Search2026.02 | 64.69 | 80.79 | 70.74 | 11.01 | |
| Iterative RPOBase Model=Llama3-8B, Training Method=Iterative RPO, Decoding Strategy=Beam Search2026.02 | 58.92 | 79.72 | 59.24 | 5.34 | |
| DPOBase Model=Llama3-8B, Training Method=DPO, Decoding Strategy=Beam Search2026.02 | 54.42 | 62.54 | 57.12 | 28.39 | |
| Continual Training (CT)Base Model=Llama3-8B, Training Method=Continual Training, Decoding Strategy=Beam Search2026.02 | 53.16 | 89.35 | 63.7 | 18.33 | |
| PretrainRLBase Model=Qwen3-8B, Training Method=PretrainRL, Decoding Strategy=Beam Search2026.02 | 47.58 | 77.1 | 55.98 | 11.1 | |
| PretrainRLBase Model=Qwen3-4B, Training Method=PretrainRL, Decoding Strategy=Beam Search2026.02 | 46.2 | 78.98 | 54.11 | 7.16 | |
| DPOBase Model=Qwen3-8B, Training Method=DPO, Decoding Strategy=Beam Search2026.02 | 38.51 | 49.49 | 41.7 | 51.41 | |
| Continual Training (CT)Base Model=Qwen3-4B, Training Method=Continual Training, Decoding Strategy=Beam Search2026.02 | 37.42 | 79.32 | 45.89 | 2.57 | |
| DPOBase Model=Qwen3-4B, Training Method=DPO, Decoding Strategy=Beam Search2026.02 | 33.84 | 47.46 | 37.54 | 18.42 | |
| Iterative RPOBase Model=Qwen3-8B, Training Method=Iterative RPO, Decoding Strategy=Beam Search2026.02 | 33.5 | 75.13 | 44.71 | 1.98 | |
| Continual Training (CT)Base Model=Qwen3-8B, Training Method=Continual Training, Decoding Strategy=Beam Search2026.02 | 31.45 | 78.78 | 41.47 | 4.51 | |
| Iterative RPOBase Model=Qwen3-4B, Training Method=Iterative RPO, Decoding Strategy=Beam Search2026.02 | 30.23 | 72.23 | 41.63 | 1.69 | |
| Prompt (CoT)Base Model=Llama3-8B, Training Method=Prompting, Decoding Strategy=Chain-of-Thought2026.02 | 27.71 | 73.72 | 37.77 | 6.15 | |
| Llama3-8BBase Model=Llama3-8B, Training Method=Base Model, Decoding Strategy=Beam Search2026.02 | 27.6 | 74.14 | 38.43 | 7.54 | |
| Qwen3-8BBase Model=Qwen3-8B, Training Method=Base Model, Decoding Strategy=Beam Search2026.02 | 23.96 | 70.43 | 34.81 | 5.51 | |
| Prompt (CoT)Base Model=Qwen3-8B, Training Method=Prompting, Decoding Strategy=Chain-of-Thought2026.02 | 23.96 | 70.43 | 34.81 | 5.51 | |
| Qwen3-4BBase Model=Qwen3-4B, Training Method=Base Model, Decoding Strategy=Beam Search2026.02 | 20.65 | 66.96 | 31.49 | 4.44 | |
| Prompt (CoT)Base Model=Qwen3-4B, Training Method=Prompting, Decoding Strategy=Chain-of-Thought2026.02 | 20.02 | 65.97 | 30.12 | 4.24 |