safetensors 文件的 metadata 字段可能保存了模型的训练数据，包括学习率、数据集重复次数等参数，使用本脚本可以直接从 <a target="_blank" rel="ugc" href="http://civarchive.com">civitai.com</a> 的模型页面导出模型的 metadata。This is a simple script that allows you to export the metadata of .safetensors files on <a target="_blank" rel="ugc" href="http://civarchive.com">civitai.com</a>. The metadata may include some training parameters such as learning rate and dataset repetition etc and you can use these parameters for your project or anywhere you want.使用方法 / Usage1. 安装油猴扩展 / Install Tampermonkey<a target="_blank" rel="ugc" href="https://www.tampermonkey.net/">https://www.tampermonkey.net/</a>2. 从 GreasyFork 或者 Github 安装脚本 / Install my script from GreasyFork or Github<a target="_blank" rel="ugc" href="https://greasyfork.org/zh-CN/scripts/467975-exportcivitaimetadata">[GreasyFork]</a><a target="_blank" rel="ugc" href="https://raw.githubusercontent.com/magicFeirl/ExportCivitAIMetadata/main/index.js">[Github]</a>3. 刷新网页，进入一个模型页面 / Refresh the page and go to a model page在按钮旁边应该有个复制按钮 / You can see there is a copying button alongside the download button<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/8522bd31-5f01-4e8a-940e-7f5b90c2d663/width=525/8522bd31-5f01-4e8a-940e-7f5b90c2d663.jpeg" />4. 点击该按钮即可导出模型的 metadata / Click the copying button and then you can export the metadata of the model<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/06535416-65ef-4267-9efa-9bd315cb4ff1/width=525/06535416-65ef-4267-9efa-9bd315cb4ff1.jpeg" />5. metadata 中可能有用的 keys / useful keys of metadata<ul><li>"ss_learning_rate": 学习率，控制模型训练时权重更新的速度。</li><li>"ss_num_epochs": 训练的总 epoch 数，指定模型将对数据集进行多少次完整的训练循环。</li><li>"ss_num_train_images": 训练图像数，指定用于训练的图像数量。</li><li>"ss_text_encoder_lr": 文本编码器学习率，控制文本编码器网络在训练过程中的权重更新速度。</li><li>and more!</li></ul>如果你觉得这个脚本对你有帮助的话可以点个 star / Please give me a star if you think this script is useful :).<a target="_blank" rel="ugc" href="https://github.com/magicFeirl/ExportCivitAIMetadata">👉[Github]</a>从其它网站读取 .safetensors 文件的 metadata见 SafetensorsHeaderReader 类

Files with SHA256: 1ebc3d6e07c59eb0e1aab0dfef98420383b4058f6616e0f4dff3589dc54a42bd