EoRA: Fine-tuning-free Compensation for Compressed LLM with Eigenspace Low-Rank Approximation - View it on GitHub