LLaMA 33B
Meta · 2023-02-24 · 32.5B parameters
LLaMA 33B is the 32.5-billion-parameter pretrained base model in Meta's original LLaMA family, publicly announced on February 24, 2023. Meta trained it on a reported 1.4 trillion tokens with a 2,048-token context window and released access to the weights case-by-case under a non-commercial bespoke research license.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 103.6 |
| Epoch Capabilities Index (ECI) | 106 |
| HellaSwag (Unspecified Scoring Protocol) | 82.8 |
| LAMBADA | 77.2 |
| OpenBookQA | 58.6 |
| WinoGrande | 76.0 |