引导式测验 · 中等 等级

GLIDE 扩散模型测验

探索 GLIDE 文本引导生成和修复、条件训练、无分类器采样及其历史模型架构。

相关引导路径滑翔扩散模型
问题 1 的 8

GLIDE 发现哪种引导方法可以生成更好、更迅速、更忠实的图像？

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。

  1. GLIDE 发现哪种引导方法可以生成更好、更迅速、更忠实的图像？
    • 无分类器指导
    • 剪辑指导
    • 无指导
    • 手动编辑
  2. What size did the published full GLIDE base diffusion model have, before counting its upsampler?
    • About 3.5 thousand parameters
    • About 350 billion parameters
    • About 3.5 billion parameters
    • No trainable parameters
  3. 训练期间无分类器指导如何工作？
    • 模型只会看到正确的提示
    • 同时训练一个单独的图像分类器
    • 提示被随机转换为数字
    • 模型有时会看到真正的提示，有时会看到空白提示
  4. Which later system describes its diffusion decoder as modifying the GLIDE architecture in the DALL-E 2 research paper?
    • The original text-only GPT-2 model
    • The unCLIP system associated with DALL-E 2
    • The Whisper speech recognizer
    • The CLIP contrastive encoder alone
  5. 为什么 OpenAI 只公开发布一个较小的、经过过滤的 GLIDE 版本？
    • 完整模型丢失
    • 它只在超级计算机上运行
    • 对滥用强大图像生成器的担忧
    • 来自第三方的许可
  6. GLIDE 支持哪些实用的编辑功能？
    • 文本驱动修复
    • 音频转录
    • 电子表格公式
    • 视频渲染
  7. How did GLIDE's upsampler complement its base text-conditioned diffusion model?
    • It replaced the text prompt with an audio transcript
    • It increased the resolution of the image produced by the base model
    • It converted the image into the training tokenizer's vocabulary
    • It supplied manual labels for every generated object
  8. At GLIDE sampling time, how does classifier-free guidance strengthen prompt conditioning?
    • It retrains the model weights after every generated pixel
    • It requires a separate class-label classifier to score every step
    • It replaces the conditional prediction with the unconditional one
    • It combines conditional and unconditional predictions, extrapolating toward the conditional one