Medical Reasoning on Medical Reasoning Evaluation Suite m23k
57.5Average PerformanceTEMP
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TEMPBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 57.5 | 59.6 | 74.2 | 76.2 | 62.7 | 47.8 | 61 | 59.4 | 55.4 | 18.3 | 60.5 | |
| TEMPBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 56.5 | 58.7 | 72.3 | 76 | 63.8 | 46.2 | 61.4 | 59.4 | 50.2 | 16.1 | 61.4 | |
| TEMPBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 56.5 | 58.7 | 72.3 | 76 | 63.8 | 46.2 | 61.4 | 59.4 | 50.2 | 16.1 | 61.4 | |
| TEMPBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 55.9 | 58.1 | 68.3 | 75.2 | 63.2 | 48.2 | 61.4 | 57.5 | 47.9 | 15.9 | 63.3 | |
| S2LBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 55.8 | 58.4 | 71.2 | 75.4 | 61.9 | 46.8 | 59.5 | 55.7 | 51.1 | 17.8 | 60.4 | |
| Middle PerplexityBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 55.1 | 57.9 | 68.9 | 70.8 | 60.1 | 47.3 | 59.8 | 55.7 | 54.9 | 17.6 | 58.2 | |
| RandomBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 55.1 | 57.3 | 68.6 | 75.1 | 62.6 | 45.4 | 62.4 | 56 | 48.7 | 15.7 | 59.5 | |
| RandomBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 55.1 | 57.3 | 68.6 | 75.1 | 62.6 | 45.4 | 62.4 | 56 | 48.7 | 15.7 | 59.5 | |
| Middle PerplexityBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 54.8 | 57.9 | 68.5 | 74.5 | 62.8 | 44.6 | 60.2 | 54.1 | 49.4 | 15.4 | 60.4 | |
| S2LBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 54.8 | 57.4 | 68.6 | 75 | 61.8 | 46 | 61.9 | 53.9 | 47.7 | 14.7 | 61.1 | |
| Middle PerplexityBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 54.8 | 57.9 | 68.5 | 74.5 | 62.8 | 44.6 | 60.2 | 54.1 | 49.4 | 15.4 | 60.4 | |
| S2LBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 54.8 | 57.4 | 68.6 | 75 | 61.8 | 46 | 61.9 | 53.9 | 47.7 | 14.7 | 61.1 | |
| RandomBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 54.7 | 58.5 | 70 | 63.2 | 61.5 | 46.2 | 61.3 | 57.8 | 52.3 | 18.3 | 58.1 | |
| RandomBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 54.4 | 58.1 | 67.3 | 77 | 59.3 | 46.4 | 61.3 | 54.9 | 44.3 | 14.7 | 60.5 | |
| Middle PerplexityBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 54.4 | 57.6 | 67.5 | 74.3 | 62.3 | 45.8 | 59.1 | 54.4 | 46.4 | 16.1 | 60.6 | |
| S2LBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 54.4 | 57.3 | 68.7 | 73.7 | 58.4 | 46.4 | 62.1 | 55 | 46.8 | 14.8 | 61.1 | |
| LearnabilityBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 54.3 | 57.8 | 68.1 | 70.5 | 58.4 | 45.7 | 60.3 | 53.5 | 51.8 | 18.1 | 58.5 | |
| Embedding DiversityBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 54.3 | 57.4 | 66.7 | 74.8 | 61.7 | 45.3 | 61.9 | 52.1 | 47.9 | 15.3 | 60.3 | |
| Embedding DiversityBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 54.3 | 57.4 | 66.7 | 74.8 | 61.7 | 45.3 | 61.9 | 52.1 | 47.9 | 15.3 | 60.3 | |
| Embedding DiversityBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 54.2 | 58.2 | 70.1 | 61.2 | 61.6 | 45 | 61 | 56.7 | 52.4 | 17.6 | 58.4 | |
| Embedding DiversityBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 54 | 57.6 | 67.6 | 74.6 | 59.1 | 46.8 | 61 | 52.3 | 46.6 | 14.4 | 60.4 | |
| LearnabilityBackbone=Qwen2.5-7B-Instruct, Data Selection Budget=500, Evaluation Protocol=Fine-tuning2026.06 | 53.7 | 56.5 | 64.9 | 73.7 | 60.2 | 46.8 | 60.6 | 52.8 | 48.1 | 14.1 | 59.3 | |
| LearnabilityBase Model=Qwen2.5-7B-Instruct, Number of training examples=5002026.06 | 53.7 | 56.5 | 64.9 | 73.7 | 60.2 | 46.8 | 60.6 | 52.8 | 48.1 | 14.1 | 59.3 | |
| LearnabilityBase Model=Qwen2.5-7B-Instruct, Number of training examples=2502026.06 | 53.6 | 56.1 | 65.6 | 73.1 | 58.7 | 47.5 | 61.7 | 52.6 | 47.2 | 13.4 | 60 | |
| m1kBase Model=Llama-3.1-8B-Instruct, Evaluation Protocol=Finetuning, Sample Size=1000 curated examples2026.06 | 53.1 | 58.4 | 74.2 | 32 | 63 | 46.5 | 60.7 | 61.5 | 55 | 17.3 | 62.5 |