引導式測驗 · 硬 等級

YaRN 上下文視窗縮放測驗

查看 YaRN 頻段、過渡混合、注意力溫度、適配成本以及支援的長度和性能之間的差異。

相關引導路徑紗線上下文視窗縮放
問題 1 的 8

YaRN 修改了什麼位置編碼方案來擴展上下文長度?

預覽每個問題8 問題

當您參加測驗時,就會出現答案和解釋。

  1. YaRN 修改了什麼位置編碼方案來擴展上下文長度?
    • 旋轉位置嵌入 (RoPE)
    • 絕對是正弦嵌入
    • 學習位置查找表
    • ALiBi 線性偏差
  2. YaRN 如何處理高頻與低頻 RoPE 維度?
    • 它相等地插值兩者
    • 它完全消除了高頻維度
    • 它只修改高頻維度
    • 它使高頻維度基本保持不變並插入低頻維度
  3. How should YaRN's reported data and fine-tuning efficiency be interpreted?
    • As proof that no YaRN model ever needs training
    • As results for its evaluated model-extension setups, rather than a fixed guarantee for every model
    • As a universal promise that every dataset size works equally well
    • As removal of the need for GPUs or inference computation
  4. 除了重新調整頻率之外,YaRN 還應用哪些額外調整來維持遠端注意力穩定?
    • 輟學率增加
    • 詞彙量更大
    • 注意溫度(softmax 縮放)調整
    • 漸層裁切
  5. What motivates evaluating a RoPE model before using a context far beyond its original training length?
    • The model necessarily becomes perfectly accurate at longer lengths
    • The context extension automatically reduces memory to zero
    • A larger window proves every passage is equally easy to retrieve
    • Positional and attention behavior may degrade outside the conditions it learned
  6. 樸素的線性位置插值會損害 YaRN 旨在保留的哪些功能?
    • 附近標記之間良好的局部關係
    • 模型的詞彙量大小
    • 分詞器的速度
    • 層數
  7. What happens in the transition region between YaRN's high- and low-frequency treatments?
    • Every dimension is assigned a random vocabulary token
    • All transition dimensions are deleted from attention
    • The scheme blends original and interpolated frequency treatment across the relevant dimensions
    • One identical scaling rule ignores frequency differences entirely
  8. What does a checkpoint's advertised longer YaRN context limit fail to establish by itself?
    • Whether the positional scheme can accept any coordinates at all
    • Reliable performance for every intended task and position within that window
    • The spelling of the model's name
    • A requirement that every inference request retrain all weights