Parameter-efficient fine-tuning for ChatGLM-6B using P-Tuning v2 and LoRA. Reduces trainable parameters by 99% while maintaining performance. Supports single GPU training.