LLaMA-65B
Code source ouvertpar Meta · Sorti le 2024-01-01
61.9
score moyen
N/A
Prix d'entrée
N/A
Prix de sortie
N/A
Fenêtre de contexte
text
Type
Tested on 10 benchmarks with 61.9% average. Top scores: TriviaQA (86.0%), HellaSwag (78.9%), LAMBADA (77.7%).
Scores de benchmark
| Benchmark | Catégorie | Score | Bar |
|---|---|---|---|
| TriviaQA | knowledge | 86.0 | |
| HellaSwag | knowledge | 78.9 | |
| LAMBADA | knowledge | 77.7 | |
| PIQA | knowledge | 65.6 | |
| ARC AI2 | knowledge | 59.3 | |
| GSM8K | math | 54.4 | |
| Winogrande | knowledge | 54.0 | |
| MMLU | knowledge | 51.2 | |
| OpenBookQA | knowledge | 46.9 | |
| BBH | reasoning | 44.5 |