LLaMA-13B
Code source ouvertpar Meta · Sorti le 2024-01-01
45.7
score moyen
N/A
Prix d'entrée
N/A
Prix de sortie
N/A
Fenêtre de contexte
text
Type
Tested on 11 benchmarks with 45.7% average. Top scores: TriviaQA (77.9%), LAMBADA (75.2%), HellaSwag (72.3%).
Scores de benchmark
| Benchmark | Catégorie | Score | Bar |
|---|---|---|---|
| TriviaQA | knowledge | 77.9 | |
| LAMBADA | knowledge | 75.2 | |
| HellaSwag | knowledge | 72.3 | |
| PIQA | knowledge | 60.2 | |
| Winogrande | knowledge | 46.0 | |
| OpenBookQA | knowledge | 41.9 | |
| ARC AI2 | knowledge | 36.9 | |
| MMLU | knowledge | 30.3 | |
| ScienceQA | knowledge | 24.4 | |
| GSM8K | math | 20.6 | |
| BBH | reasoning | 17.2 |