Question Answering on CondaQA (val)
77.5AccuracyLlama 3.1 70B
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Llama 3.1 70BLearning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=70B2026.04 | 77.5 | 43.3 | 83.7 | 61.8 | 54.9 | |
| GPT-4oLearning Paradigm=In-context learning, Few-shot=Few-shot2026.04 | 72.9 | 34.4 | 78.7 | 52.6 | 47.9 | |
| Llama 3.1 8B + Best of (¬ATOMIC, ¬ANION)Learning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=8B, Pre-training Corpora=¬ATOMIC or ¬ANION2026.04 | 71.5 | 33.5 | 72.9 | 48.7 | 46.4 | |
| Qwen2 7B + Best of (¬ATOMIC, ¬ANION)Learning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=7B, Pre-training Corpora=¬ATOMIC or ¬ANION2026.04 | 69.7 | 32.7 | 71.2 | 46.3 | 45.5 | |
| Llama 3.1 8BLearning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=8B2026.04 | 68.7 | 31.5 | 69 | 48.1 | 44.4 | |
| RoBERTa-large + Best of (¬ATOMIC, ¬ANION)Learning Paradigm=Fully supervised, # Params.=355M, Pre-training Corpora=¬ATOMIC or ¬ANION2026.04 | 68.5 | 34.3 | 66.4 | 47.6 | 50.1 | |
| Llama 3.1 8B + ATOMIC + ANIONLearning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=8B, Pre-training Corpora=ATOMIC + ANION2026.04 | 67.6 | 27.5 | 69.5 | 44.3 | 41.4 | |
| RoBERTa-large + ATOMIC + ANIONLearning Paradigm=Fully supervised, # Params.=355M, Pre-training Corpora=ATOMIC + ANION2026.04 | 67 | 32.9 | 65.6 | 46.3 | 48.1 | |
| UnifiedQA-v2-largeLearning Paradigm=Fully supervised, # Params.=770M2026.04 | 66.7 | 30.2 | 64 | 43.7 | 46.5 | |
| InstructGPT + COTLearning Paradigm=In-context learning, Few-shot=Few-shot2026.04 | 66.3 | 27.3 | 64.2 | 45.1 | 44.9 | |
| OpenAI o1Learning Paradigm=In-context learning, Few-shot=Zero-shot2026.04 | 65.3 | 24.9 | 67.4 | 43.8 | 38.6 | |
| Qwen2 7BLearning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=7B2026.04 | 65.1 | 24.9 | 65.9 | 39.7 | 39.2 | |
| RoBERTa-largeLearning Paradigm=Fully supervised, # Params.=355M2026.04 | 64.9 | 29.6 | 61.9 | 41.4 | 45.8 | |
| Qwen2 7B + ATOMIC + ANIONLearning Paradigm=In-context learning, Few-shot=Few-shot, # Params.=7B, Pre-training Corpora=ATOMIC + ANION2026.04 | 64.1 | 22.8 | 65.2 | 38.4 | 37.7 |