Meta Llama 3 8B Instruct
Código abertopor Meta · Lançado em 2024-04-17
45.2
pontuação média
N/A
Preço de entrada
N/A
Preço de saída
N/A
Janela de contexto
text-generation
Tipo
Tested on 25 benchmarks with 45.2% average. Top scores: JSQuAD (89.5%), JCommonsenseQA (87.7%), IFEval (74.1%).
Pontuações de benchmark
| Benchmark | Categoria | Pontuação | Bar |
|---|---|---|---|
| JSQuAD | language | 89.5 | |
| JCommonsenseQA | language | 87.7 | |
| IFEval | language | 74.1 | |
| JNLI | language | 61.1 | |
| MMMLU — Spanish | language | 55.8 | |
| MMMLU — French | language | 55.8 | |
| MMMLU — Portuguese | language | 55.5 | |
| MMMLU — German | language | 53.5 | |
| MMMLU — Italian | language | 53.3 | |
| MMMLU — Chinese | language | 51.4 | |
| MMMLU — Indonesian | language | 51.0 | |
| LLM-JP — Overall | language | 49.6 | |
| JMMLU | language | 46.7 | |
| MMMLU — Korean | language | 46.5 | |
| MMMLU — Japanese | language | 42.3 | |
| MMMLU — Hindi | language | 41.4 | |
| MMMLU — Arabic | language | 40.5 | |
| MMMLU — Swahili | language | 37.5 | |
| MMMLU — Bengali | language | 36.4 | |
| MMMLU — Yoruba | language | 31.0 | |
| MMLU-PRO | knowledge | 29.6 | |
| BBH (HuggingFace) | general | 28.2 | |
| MATH Level 5 | math | 8.7 | |
| MUSR | reasoning | 1.6 | |
| GPQA | knowledge | 1.2 |
