Qwen2 7B Instruct vs Meta Llama 3 8B Instruct
Lado a lado. Cada métrica. Cada benchmark.
| Tipo | Qwen2 7B Instruct | Meta Llama 3 8B Instruct |
|---|---|---|
| Provider |  Alibaba |  Meta |
| puntuación promedio | 50.5 | 45.2 |
| Precio de entrada | - | - |
| Precio de salida | - | - |
| Ventana de contexto | - | - |
| Publicado el | 2024-06-04 | 2024-04-17 |
| Código abierto | Open Source | Open Source |
Puntuaciones de benchmark
25 benchmarks · Qwen2 7B Instruct: 22, Meta Llama 3 8B Instruct: 3
| Benchmark | Categoría | Qwen2 7B Instruct | Meta Llama 3 8B Instruct |
|---|---|---|---|
| BBH (HuggingFace) | general | 37.8 | 28.2 |
| GPQA | knowledge | 6.4 | 1.2 |
| IFEval | language | 56.8 | 74.1 |
| JCommonsenseQA | language | 89.1 | 87.7 |
| JMMLU | language | 56.5 | 46.7 |
| JNLI | language | 81.3 | 61.1 |
| JSQuAD | language | 89.6 | 89.5 |
| LLM-JP — Overall | language | 51.7 | 49.6 |
| MATH Level 5 | math | 27.6 | 8.7 |
| MMLU-PRO | knowledge | 31.6 | 29.6 |
| MMMLU — Arabic | language | 50.7 | 40.5 |
| MMMLU — Bengali | language | 43.4 | 36.4 |
| MMMLU — Chinese | language | 61.8 | 51.4 |
| MMMLU — French | language | 60.8 | 55.8 |
| MMMLU — German | language | 57.1 | 53.5 |
| MMMLU — Hindi | language | 45.1 | 41.4 |
| MMMLU — Indonesian | language | 54.1 | 51.0 |
| MMMLU — Italian | language | 59.0 | 53.3 |
| MMMLU — Japanese | language | 56.6 | 42.3 |
| MMMLU — Korean | language | 54.0 | 46.5 |
| MMMLU — Portuguese | language | 60.1 | 55.5 |
| MMMLU — Spanish | language | 60.2 | 55.8 |
| MMMLU — Swahili | language | 34.3 | 37.5 |
| MMMLU — Yoruba | language | 30.2 | 31.0 |
| MUSR | reasoning | 7.4 | 1.6 |