← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8664.a · 生成模型:GPT-6

配置开关须按指定版本验证

理解等级:高级

示例所锁Transformers4.35.0使用load_in_4bit启用四位加载,训练轮数字段为num_train_epochs。书中的use_4bit与max_train_epochs不能按名字相似就假设生效;构造成功也应核对实际加载精度及可训练参数。

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

      use_4bit = True bnb_4bit_compute_dtype = 'float16' bnb_4bit_quant_type = 'nf4' use_nested_quant = False Finally, we use the Transformer Reinforcement Learning (TRL) library that, in addition to reinforcement learning, provides support for supervised fine-tuning. Here are some recommended default values: max_seq_length = 128 # Packing is used to place multiple instructions in the same input sequence packing = True Putting It All Together Now that we have

    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:摘录中关于配置开关须按指定版本验证的定义、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:use_4bit = True bnb_4bit_compute_dtype = 'float16' bnb_4bit_quant_type = 'nf4' use_nested_quant = False Finally, we use the Transformer Reinforcement Learning (TRL) library that, in addition to reinforcement learning, provides support for supervised fine-tuning. Here are some recommended default values: max_seq_length = 128 # Packing is used to place multiple instructions in the same input sequence packing = True Putting It All Together Now that we have