引导式测验 · 中等 等级
Kimi 和 Moonshot 人工智能测验
回顾 Kimi 的长上下文起源、Kimi K2 的专家混合设计，以及上下文和模型功能为何因版本而异。
问题 1 的 8
Moonshot 在 2024 年 1 月公开测试的 moonshot-v1 API 宣布的最大上下文规模是多少？
预览每个问题8 问题
当您参加测验时，就会出现答案和解释。
- Moonshot 在 2024 年 1 月公开测试的 moonshot-v1 API 宣布的最大上下文规模是多少？
- Up to 128K tokens
- Up to 8K tokens
- Up to 32K tokens
- Up to 1M tokens
- When does Moonshot AI’s official timeline say the company was founded?
- Spring 2018
- Spring 2023
- Fall 2020
- Winter 2025
- 阿里巴巴在其 FY2024 年度报告中披露的主要投资者是谁？
- OpenAI, with an approximately 36% equity interest
- Anthropic, with an approximately 36% equity interest
- Moonshot AI, with an approximately 36% equity interest reported for that fiscal year
- DeepMind, with an approximately 36% equity interest
- 后来的 Kimi 模型（例如 K2）使用什么架构来保持大规模的效率？
- 纯循环网络
- 决策树
- 支持向量机
- 混合专家 (MoE)
- 2025 年初发布的 Kimi k1.5 旨在改进什么？
- 逐步推理
- 图像放大
- 电池效率
- 3D渲染
- 为什么长上下文窗口实际上很有用？
- 它减少了模型的参数数量
- 它让模型一次考虑整个长文档
- 它消除了对训练数据的需要
- 它保证了事实的准确性
- What scale does the Kimi K2 technical report give for total and activated parameters?
- 70 billion total, with all 70 billion activated
- 7 billion total, with 32 billion activated
- 1 trillion total parameters, with 32 billion activated
- 1 trillion total, with all parameters activated for every token
- Why should a comparison of Kimi model capabilities identify the exact release?
- Every release has identical capabilities
- The product name alone establishes the model’s benchmark score
- Version identifiers matter only for billing
- Context limits, modalities, and tool behavior can differ across model versions