Question Answering on TriviaQA (Acc., Steps, Speed Up)
79.89AccuracySandwich
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SandwichBackbone=Llama3.1-8B-Instruct2025.04 | 79.89 | — | — | |
| InstructionalBackbone=Llama3.1-8B-Instruct2025.04 | 79.44 | — | — | |
| Robustness via ReferencingBackbone=Llama3.1-8B-Instruct2025.04 | 79.44 | — | — | |
| NoneBackbone=Llama3.1-8B-Instruct2025.04 | 79.11 | — | — | |
| ReminderBackbone=Llama3.1-8B-Instruct2025.04 | 78.89 | — | — | |
| Robustness via ReferencingBackbone=Qwen2-7B-Instruct2025.04 | 78 | — | — | |
| SandwichBackbone=Llama3-8B-Instruct2025.04 | 77.44 | — | — | |
| Robustness via ReferencingBackbone=Llama3-8B-Instruct2025.04 | 77.44 | — | — | |
| SandwichBackbone=Qwen2-7B-Instruct2025.04 | 77.44 | — | — | |
| SpotlightBackbone=Llama3.1-8B-Instruct2025.04 | 77.44 | — | — | |
| NoneBackbone=Qwen2-7B-Instruct2025.04 | 77.22 | — | — | |
| SpotlightBackbone=Qwen2-7B-Instruct2025.04 | 77.11 | — | — | |
| ReminderBackbone=Qwen2-7B-Instruct2025.04 | 76.78 | — | — | |
| InstructionalBackbone=Qwen2-7B-Instruct2025.04 | 76.22 | — | — | |
| StruQBackbone=Llama3.1-8B-Instruct2025.04 | 76.22 | — | — | |
| NoneBackbone=Llama3-8B-Instruct2025.04 | 75.78 | — | — | |
| ReminderBackbone=Llama3-8B-Instruct2025.04 | 75.67 | — | — | |
| StruQBackbone=Llama3-8B-Instruct2025.04 | 75.56 | — | — | |
| StruQBackbone=Qwen2-7B-Instruct2025.04 | 75 | — | — | |
| SpotlightBackbone=Llama3-8B-Instruct2025.04 | 74.22 | — | — | |
| InstructionalBackbone=Llama3-8B-Instruct2025.04 | 73.89 | — | — | |
| RLBackbone=Llama-3.1-8B2026.05 | 69.89 | — | — | |
| RLBackbone=OLMo-2-7B2026.05 | 65.85 | — | — | |
| RFTBackbone=Llama-3.1-8B2026.05 | 65.21 | — | — | |
| DPOBackbone=Llama-3.1-8B2026.05 | 64.4 | — | — | |
| BaseBackbone=Llama-3.1-8B2026.05 | 63.54 | — | — | |
| SFTBackbone=Llama-3.1-8B2026.05 | 62.35 | — | — | |
| RLBackbone=Qwen2.5-7B2026.05 | 59.94 | — | — | |
| RFTBackbone=OLMo-2-7B2026.05 | 59.63 | — | — | |
| SFTBackbone=OLMo-2-7B2026.05 | 57.67 | — | — | |
| DPOBackbone=OLMo-2-7B2026.05 | 55.62 | — | — | |
| BaseBackbone=OLMo-2-7B2026.05 | 54.86 | — | — | |
| RFTBackbone=Qwen2.5-7B2026.05 | 53.65 | — | — | |
| OriginalModel=LLaDA, Decoding Strategy=Step-by-step2026.03 | 52.64 | 256 | — | |
| DPOBackbone=Qwen2.5-7B2026.05 | 52.43 | — | — | |
| KLASSModel=LLaDA, Decoding Strategy=Parallel2026.03 | 52.2 | 177.4 | 1.44 | |
| Fast-dLLMModel=LLaDA, Decoding Strategy=Parallel2026.03 | 52.12 | 124.4 | 2.06 | |
| DAPDModel=LLaDA, Decoding Strategy=Parallel2026.03 | 52.08 | 66.2 | 3.87 | |
| SFTBackbone=Qwen2.5-7B2026.05 | 51.79 | — | — | |
| EBModel=LLaDA, Decoding Strategy=Parallel2026.03 | 51.2 | 131.3 | 1.95 | |
| BaseBackbone=Qwen2.5-7B2026.05 | 49.39 | — | — |