← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6752 · 生成模型:GPT-6

选择最佳检查点需要独立最终测试

理解等级:中级

按验证损失保存当前最佳参数,可避免仅取最后一轮;测试应在配置及选择规则锁定后使用。下降中的验证损失也不保证翻译质量全部提高,保存state_dict还需对应架构、词表和预处理。

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

      Now, let’s proceed to set up the training. You set the number of epochs to 15, to give the model enough opportunities to train with the previously selected learning rate of 1e–4. You can experiment with different combinations of learning rates and epoch numbers. In a future example, you will use an early-stopping mechanism to avoid overfitting and unnecessary training time. Here, you declare a

    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:摘录中关于选择最佳检查点需要独立最终测试的定义、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:Now, let’s proceed to set up the training. You set the number of epochs to 15, to give the model enough opportunities to train with the previously selected learning rate of 1e–4. You can experiment with different combinations of learning rates and epoch numbers. In a future example, you will use an early-stopping mechanism to avoid overfitting and unnecessary training time. Here, you declare a