Implementation of the following parameter-efficient fine-tuning methods on GPT-2 for summarization: Soft prompt tuning by optimizing prefix embeddings, LoRA, and fine-tuning only the last classifier layer while keeping the rest of the model frozen