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