知识卡片

普通读书笔记卡 · 8664.d · 生成模型:GPT-6

保存适配器不同于保存完整模型

理解等级:中级

PEFT训练保存的制品可能只含适配器增量,加载还需匹配基础模型、配置与tokenizer。目录扩展名不决定文件语义;完整续训还可能需要优化器、调度器和随机状态,不能把save_pretrained一概称作完整恢复。

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

      sft.model.save_pretrained('/path/to/llama-2-it.csv') The relationship between the hyperparameters is very complex, and you might find surprising results. It will take several iterations before you hit the sweet spot. However, do not spend too much time squeezing out the last bit of performance from your fine-tuning, as that time is better spent developing better training data. In the next section, we will learn how to create effective training

    参考来源

    - 位置:《Designing Large Language Model Applications》第06章《微调》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch06.html) - 结论依据:摘录中关于保存适配器不同于保存完整模型的定义、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:sft.model.save_pretrained('/path/to/llama-2-it.csv') The relationship between the hyperparameters is very complex, and you might find surprising results. It will take several iterations before you hit the sweet spot. However, do not spend too much time squeezing out the last bit of performance from your fine-tuning, as that time is better spent developing better training data. In the next section, we will learn how to create effective training