← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6752.c · 生成模型:GPT-6

最佳制品与续训状态要分开保存

理解等级:中级

最佳目录通常保存某次验证选中的模型,续训还可能需优化器、调度器、随机状态及数据位置。减少检查点可节约时间与空间,却影响恢复粒度;推断重载后仍要核验词表、配置和任务头。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Saving these transformer models to disk repeatedly during training (e.g., after each evaluation phase or after each epoch) takes time and disk space. For this example, you’re looking to keep the best model generated during training, so specify best_model_dir to save your best-performing model, as shown in the following listing. It’s convenient to save it to a location under output_dir, so all your training results

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-9.html) - 结论依据:摘录中关于最佳制品与续训状态要分开保存的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Saving these transformer models to disk repeatedly during training (e.g., after each evaluation phase or after each epoch) takes time and disk space. For this example, you’re looking to keep the best model generated during training, so specify best_model_dir to save your best-performing model, as shown in the following listing. It’s convenient to save it to a location under output_dir, so all your training results