GPT-4: OpenLLaMA, an open-source reproduction of Meta AI’s LLaMA model, has been released, providing researchers with a permissively licensed large language model. Trained on the RedPajama dataset with 200 billion tokens, OpenLLaMA exhibits comparable or better performance than the original LLaMA and GPT-J across most tasks. The release includes PyTorch and Jax weights, evaluation results, and a comparison against the original LLaMA models, making it a valuable resource for machine learning researchers who face challenges accessing proprietary models.
Read more at MarkTechPost…