LLaMA 7B
Meta · 2023-02-24 · 6.7B parameters
LLaMA 7B is Meta's original pretrained base language model and the smallest checkpoint in the first LLaMA family, publicly announced on February 24, 2023. Meta reported the model at 6.7 billion parameters and 1.0 trillion training tokens and distributed its weights to approved researchers under a non-commercial bespoke license.
Benchmark scores
| Benchmark | Score |
|---|---|
| Capability | 84.1 |
| Epoch Capabilities Index (ECI) | 95.2 |
| LAMBADA | 73.3 |
| ScienceQA | 36.2 |