DeepSeek R1 Distill Llama 8B
Código abiertopor DeepSeek · Publicado el 2025-01-20
33.6
puntuación promedio
N/A
Precio de entrada
N/A
Precio de salida
N/A
Ventana de contexto
text-generation
Tipo
Tested on 11 benchmarks with 33.6% average. Top scores: JSQuAD (80.2%), JNLI (69.4%), JCommonsenseQA (62.4%).
Puntuaciones de benchmark
| Benchmark | Categoría | Puntuación | Bar |
|---|---|---|---|
| JSQuAD | language | 80.2 | |
| JNLI | language | 69.4 | |
| JCommonsenseQA | language | 62.4 | |
| LLM-JP — Overall | language | 41.4 | |
| IFEval | language | 37.8 | |
| JMMLU | language | 37.8 | |
| MATH Level 5 | math | 22.0 | |
| MMLU-PRO | knowledge | 12.1 | |
| BBH (HuggingFace) | general | 5.3 | |
| GPQA | knowledge | 0.7 | |
| MUSR | reasoning | 0.5 |
