📄 中文摘要
大语言模型（LLMs）虽然强大，但其规模庞大，像GPT风格的变换器模型包含数十亿个参数，运行这些模型需要昂贵的GPU、高内存和强大的计算能力。然而，许多参数是冗余的，这为低秩矩阵分解提供了机会。低秩矩阵分解可以有效地减少模型的参数数量，从而在保持模型性能的同时降低计算资源的需求。通过对权重矩阵进行优化，可以实现更高效的模型运行，推动LLMs在实际应用中的普及。
低秩矩阵分解：在不破坏大语言模型的情况下缩小其规模
📄 English Summary
Low-Rank Matrix Factorization: Shrinking LLMs Without Breaking Their Brain
Large Language Models (LLMs) are powerful but also massive, with models like GPT-style transformers containing billions of parameters, requiring expensive GPUs, high memory, and significant computational power to run. However, many of these parameters are redundant, which presents an opportunity for Low-Rank Matrix Factorization. This technique can effectively reduce the number of parameters in the model, lowering the computational resource requirements while maintaining performance. By optimizing the weight matrices, more efficient model operation can be achieved, facilitating the broader application of LLMs in real-world contexts.