who push the state-of-the-art in model architectures, algorithms, and training techniques ... often fine-tuning large models into smaller, more efficient versions for anyone to use. For instance, Meta ...