📄 中文摘要
训练大型语言模型（LLM）时，显存的需求显著增加，尤其是在微调自定义数据集时。除了存储模型权重外，还需考虑优化器状态、梯度和激活内存等因素。为了避免在环境设置、下载大型.safetensors文件和编写训练脚本时遭遇显存不足的错误，建议使用专门的工具来评估显存需求，从而更高效地进行模型微调。
微调大型语言模型需要多少显存？别再猜了，使用这个工具吧。
📄 English Summary
How Much VRAM Do You Need to Fine-Tune an LLM? Stop Guessing and Use This Tool.
Training a Large Language Model (LLM) locally can lead to the frustrating 'RuntimeError: CUDA out of memory' message, especially when fine-tuning on a custom dataset. The hardware requirements increase dramatically, as one must consider not only the model weights but also optimizer states, gradients, and activation memory. To prevent hours of setup, downloading large .safetensors files, and writing training scripts only to encounter immediate crashes, utilizing a dedicated tool to assess VRAM requirements is recommended for more efficient model fine-tuning.