Large Language Model Evaluation on HuggingFace Open LLM Leaderboard (lm-eval-harness default)
84.34HellaSwagTeacher
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| TeacherDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B, Role=Teacher Model (Qwen-2.5-14B)2026.01 | 84.34 | 67.06 | 79.74 | 58.51 | 80.58 | 84.23 | 75.74 | — | — | |
| DPODistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 82.42 | 60.84 | 65.26 | 52.16 | 78.31 | 54.87 | 65.64 | — | — | |
| CTPDDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 82.25 | 63.92 | 66.65 | 55.22 | 79.29 | 57.47 | 67.42 | — | — | |
| StudentDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B, Role=Student Model (Llama-3.1-8B)2026.01 | 81.99 | 57.59 | 65.48 | 45.19 | 77.43 | 50.27 | 62.99 | — | — | |
| TIS-DPODistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 81.08 | 61.92 | 66.73 | 53.86 | 79.05 | 54.31 | 66.16 | — | — | |
| LoPTTraining Dataset=MetaMathQA, Model=Qwen2.5-7B-Ins, Peak Mem (GB)=38.4 (↓35%), Speed (tok/s)=20,511 (↑5%)2026.05 | 80.97 | — | 74.23 | 62.66 | 74.03 | 87.21 | — | 51.46 | 64.33 | |
| SFTDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 80.94 | 60.92 | 65.58 | 51.72 | 77.42 | 50.64 | 64.54 | — | — | |
| Multi-Level OTDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 80.87 | 60.93 | 65.39 | 51.99 | 77.35 | 50.95 | 64.58 | — | — | |
| E2ETraining Dataset=MetaMathQA, Model=Qwen2.5-7B-Ins, Peak Mem (GB)=58.7, Speed (tok/s)=19,5672026.05 | 80.85 | — | 74.13 | 62.63 | 71.51 | 86.8 | — | 51.76 | 63.91 | |
| TeacherDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B, Role=Teacher Model (Qwen-2.5-7B)2026.01 | 80.34 | 63.57 | 74.28 | 56.37 | 75.77 | 81.34 | 71.95 | — | — | |
| LoPTTraining Dataset=MetaMathQA, Model=Llama-3.1-8B-Ins, Peak Mem (GB)=39.9 (↓35%), Speed (tok/s)=19,194 (↑6%)2026.05 | 79.71 | — | 67.94 | 52.93 | 76.64 | 79.27 | — | 43.99 | 60.92 | |
| LoPTTraining Dataset=Alpaca-52K, Model=Qwen2.5-7B-Ins, Peak Mem (GB)=37.2 (↓36%), Speed (tok/s)=15,971 (↑3%)2026.05 | 79.46 | — | 74.08 | 54.86 | 75.63 | 79.38 | — | 41.22 | 63.96 | |
| BaseTraining Dataset=Alpaca-52K, Model=Llama-3.1-8B-Ins2026.05 | 79.42 | — | 68.2 | 54.56 | 77.5 | 69.67 | — | 44.54 | 60.49 | |
| BaseTraining Dataset=MetaMathQA, Model=Llama-3.1-8B-Ins2026.05 | 79.42 | — | 68.2 | 54.56 | 77.5 | 69.67 | — | 44.54 | 60.49 | |
| LoPTTraining Dataset=Alpaca-52K, Model=Llama-3.1-8B-Ins, Peak Mem (GB)=38.9 (↓36%), Speed (tok/s)=14,882 (↑3%)2026.05 | 79.38 | — | 68.48 | 52.67 | 76.64 | 72.1 | — | 46.03 | 60.41 | |
| ULDDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 79.36 | 57.69 | 64.96 | 50.31 | 77.66 | 50.16 | 63.35 | — | — | |
| BaseTraining Dataset=Alpaca-52K, Model=Qwen2.5-7B-Ins2026.05 | 79.35 | — | 74.17 | 64.75 | 74.51 | 83.47 | — | 57.3 | 66.72 | |
| BaseTraining Dataset=MetaMathQA, Model=Qwen2.5-7B-Ins2026.05 | 79.35 | — | 74.17 | 64.75 | 74.51 | 83.47 | — | 57.3 | 66.72 | |
| DSKDDistillation Scenario=Qwen-2.5-14B -> Llama-3.1-8B2026.01 | 79.24 | 58.19 | 64.82 | 51.77 | 74.82 | 50.11 | 63.16 | — | — | |
| E2ETraining Dataset=Alpaca-52K, Model=Qwen2.5-7B-Ins, Peak Mem (GB)=57.7, Speed (tok/s)=15,5612026.05 | 78.55 | — | 73.09 | 52.28 | 75.77 | 74.22 | — | 40.3 | 60.58 | |
| E2ETraining Dataset=Alpaca-52K, Model=Llama-3.1-8B-Ins, Peak Mem (GB)=60.6, Speed (tok/s)=14,4832026.05 | 78.31 | — | 68.67 | 47.39 | 76.09 | 46.62 | — | 30.5 | 57.59 | |
| E2ETraining Dataset=MetaMathQA, Model=Llama-3.1-8B-Ins, Peak Mem (GB)=61.4, Speed (tok/s)=18,0282026.05 | 77.7 | — | 65.29 | 52.06 | 76.56 | 76.79 | — | 34.94 | 58.7 | |
| LoPTTraining Dataset=MetaMathQA, Model=Qwen3-4B-Base, Peak Mem (GB)=25.2 (↓24%), Speed (tok/s)=26,670 (↑7%)2026.05 | 72.53 | — | 70.17 | 54.55 | 68.88 | 81.21 | — | 22.26 | 63.64 | |
| E2ETraining Dataset=MetaMathQA, Model=Qwen3-4B-Base, Peak Mem (GB)=33.0, Speed (tok/s)=25,0192026.05 | 72.43 | — | 70.18 | 54.49 | 68.69 | 81.2 | — | 20.89 | 63.57 | |
| LoPTTraining Dataset=Alpaca-52K, Model=Qwen3-4B-Base, Peak Mem (GB)=21.9 (↓ 30%), Speed (tok/s)=21,469 (↑2%)2026.05 | 69.97 | — | 69.99 | 53.91 | 67.69 | 84.62 | — | 29.39 | 61.95 | |
| E2ETraining Dataset=Alpaca-52K, Model=Qwen3-4B-Base, Peak Mem (GB)=31.5, Speed (tok/s)=21,0202026.05 | 69.8 | — | 70.12 | 53.07 | 68.19 | 79.45 | — | 25.87 | 61.77 | |
| CTPDDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 67.3 | 40.61 | 31.08 | 46.34 | 64.5 | 9.72 | 43.26 | — | — | |
| DPODistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 66.35 | 40.1 | 31.13 | 41.79 | 63.3 | 7.43 | 41.68 | — | — | |
| TIS-DPODistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 66.23 | 40.92 | 31.43 | 43.49 | 64.34 | 9.13 | 42.6 | — | — | |
| SFTDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 65.95 | 39.59 | 31.73 | 41.17 | 62.87 | 6.78 | 41.35 | — | — | |
| BaseTraining Dataset=Alpaca-52K, Model=Qwen3-4B-Base2026.05 | 65.62 | — | 70.07 | 54.85 | 67.56 | 85.14 | — | 23.84 | 57.84 | |
| BaseTraining Dataset=MetaMathQA, Model=Qwen3-4B-Base2026.05 | 65.62 | — | 70.07 | 54.85 | 67.56 | 85.14 | — | 23.84 | 57.84 | |
| StudentDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B, Role=Student Model (Llama-3.2-1B)2026.01 | 65.59 | 39.33 | 31.86 | 37.66 | 62.75 | 6.82 | 40.67 | — | — | |
| Multi-Level OTDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 65.46 | 39.76 | 31.19 | 41.73 | 63.14 | 7.12 | 41.4 | — | — | |
| ULDDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 65.09 | 40.02 | 31.15 | 41.2 | 62.77 | 5.77 | 41 | — | — | |
| DSKDDistillation Scenario=Qwen-2.5-7B -> Llama-3.2-1B2026.01 | 65.05 | 40.16 | 31.11 | 40.72 | 62.89 | 6.77 | 41.12 | — | — |