← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6502.a.5 · 生成模型:GPT-6

配置上限与能力上限分开

理解等级:中级

实现中旋转表可分配的最大位置、预训练上下文长度和实际可靠处理长度是不同概念。配置允许更长序列不保证质量保持;输入与输出共同占用位置预算,也不应把书中的典型提示/输出分配理解为不可改变的硬分区。

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

      More Precise Context Length Information For Qwen3 0.6B, the default maximum supported context length is 40,960 tokens. However, the model itself was trained with only 32,768 tokens. The 40,960-token allocation reserves 32,768 tokens for model outputs (the generated text) and 8,192 tokens for typical prompts (the user's question or instruction). To put this in perspective, 40 thousand tokens correspond to roughly one half of the

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第09章《附录C Qwen3模型源码》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/appendix-c.html) - 结论依据:摘录中关于配置上限与能力上限分开的定义、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:More Precise Context Length Information For Qwen3 0.6B, the default maximum supported context length is 40,960 tokens. However, the model itself was trained with only 32,768 tokens. The 40,960-token allocation reserves 32,768 tokens for model outputs (the generated text) and 8,192 tokens for typical prompts (the user's question or instruction). To put this in perspective, 40 thousand tokens correspond to roughly one half of the