LLaMA-7B
Code source ouvertpar Meta · Sorti le 2024-01-01
39.7
score moyen
N/A
Prix d'entrée
N/A
Prix de sortie
N/A
Fenêtre de contexte
text
Type
Tested on 11 benchmarks with 39.7% average. Top scores: LAMBADA (73.3%), TriviaQA (71.0%), HellaSwag (68.3%).
Scores de benchmark
| Benchmark | Catégorie | Score | Bar |
|---|---|---|---|
| LAMBADA | knowledge | 73.3 | |
| TriviaQA | knowledge | 71.0 | |
| HellaSwag | knowledge | 68.3 | |
| PIQA | knowledge | 59.6 | |
| OpenBookQA | knowledge | 42.9 | |
| Winogrande | knowledge | 40.2 | |
| ARC AI2 | knowledge | 30.1 | |
| ScienceQA | knowledge | 14.9 | |
| MMLU | knowledge | 14.1 | |
| BBH | reasoning | 11.3 | |
| GSM8K | math | 11.0 |