Evolution of meta's llama models and parameter-efficient fine-tuning of large language models: a survey