DeepSeek R1 Distill Qwen 14B vs Qwen2 7B Instruct
Côte à côte. Chaque métrique. Chaque benchmark.
| Type | DeepSeek R1 Distill Qwen 14B | Qwen2 7B Instruct |
|---|---|---|
| Provider |  DeepSeek |  Alibaba |
| score moyen | 56.0 | 50.5 |
| Prix d'entrée | - | - |
| Prix de sortie | - | - |
| Fenêtre de contexte | - | - |
| Sorti le | 2025-01-20 | 2024-06-04 |
| Code source ouvert | Open Source | Open Source |
Scores de benchmark
11 benchmarks · DeepSeek R1 Distill Qwen 14B: 10, Qwen2 7B Instruct: 1
| Benchmark | Catégorie | DeepSeek R1 Distill Qwen 14B | Qwen2 7B Instruct |
|---|---|---|---|
| BBH (HuggingFace) | general | 40.7 | 37.8 |
| GPQA | knowledge | 18.3 | 6.4 |
| IFEval | language | 43.8 | 56.8 |
| JCommonsenseQA | language | 93.7 | 89.1 |
| JMMLU | language | 63.4 | 56.5 |
| JNLI | language | 82.4 | 81.3 |
| JSQuAD | language | 89.8 | 89.6 |
| LLM-JP — Overall | language | 56.8 | 51.7 |
| MATH Level 5 | math | 57.0 | 27.6 |
| MMLU-PRO | knowledge | 40.7 | 31.6 |
| MUSR | reasoning | 28.7 | 7.4 |