DeepSeek R1 Distill Qwen 14B vs Qwen2 VL 7B Instruct
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | DeepSeek R1 Distill Qwen 14B | Qwen2 VL 7B Instruct |
|---|---|---|
| Provider |  DeepSeek |  Alibaba |
| pontuação média | 56.0 | 47.3 |
| Preço de entrada | - | - |
| Preço de saída | - | - |
| Janela de contexto | - | - |
| Lançado em | 2025-01-20 | 2024-08-28 |
| Código aberto | Open Source | Open Source |
Pontuações de benchmark
11 benchmarks · DeepSeek R1 Distill Qwen 14B: 9, Qwen2 VL 7B Instruct: 2
| Benchmark | Categoria | DeepSeek R1 Distill Qwen 14B | Qwen2 VL 7B Instruct |
|---|---|---|---|
| BBH (HuggingFace) | general | 40.7 | 35.9 |
| GPQA | knowledge | 18.3 | 9.3 |
| IFEval | language | 43.8 | 46.0 |
| JCommonsenseQA | language | 93.7 | 87.8 |
| JMMLU | language | 63.4 | 56.3 |
| JNLI | language | 82.4 | 74.4 |
| JSQuAD | language | 89.8 | 89.9 |
| LLM-JP — Overall | language | 56.8 | 53.0 |
| MATH Level 5 | math | 57.0 | 19.9 |
| MMLU-PRO | knowledge | 40.7 | 34.4 |
| MUSR | reasoning | 28.7 | 13.6 |