Models

← All models

OpenLLaMA 7B

OpenLM Research · 2023-06-07 · 6.7B parameters

OpenLLaMA 7B is OpenLM Research's final v1 7B base checkpoint, publicly released on June 7, 2023 as a permissively licensed reproduction of Meta's original LLaMA 7B. It follows the LLaMA 7B architecture and training hyperparameters, was trained from scratch on RedPajama in a run reported as 1T tokens, and uses a separately trained SentencePiece BPE tokenizer; its weights are licensed under Apache 2.0.

Benchmark scores