Kuis Penskalaan Jendela Konteks YaRN

Jawaban dan penjelasan muncul saat Anda mengikuti kuis.

1. Skema pengkodean posisi apa yang dimodifikasi YaRN untuk memperluas panjang konteks?
   • Penyematan Posisi Putar (Tali)
   • Penyematan sinusoidal absolut
   • Tabel pencarian posisi yang dipelajari
   • Bias linier ALiBi
2. Bagaimana YaRN menangani dimensi RoPE frekuensi tinggi dan frekuensi rendah?
   • Ini menginterpolasi keduanya secara setara
   • Ini menghilangkan seluruh dimensi frekuensi tinggi
   • Itu hanya mengubah dimensi frekuensi tinggi
   • Ini membiarkan sebagian besar dimensi frekuensi tinggi tetap utuh dan menginterpolasi dimensi frekuensi rendah
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. Selain mengubah skala frekuensi, penyesuaian tambahan apa yang diterapkan YaRN untuk menjaga kestabilan perhatian jarak jauh?
   • Peningkatan angka putus sekolah
   • Kosakata yang lebih besar
   • Penyesuaian suhu perhatian (skala softmax).
   • Kliping gradien
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. Interpolasi Posisi Linier Naif merugikan kemampuan manakah yang ingin dipertahankan oleh YaRN?
   • Hubungan lokal yang baik antara token terdekat
   • Ukuran kosakata model
   • Kecepatan tokenizer
   • Jumlah lapisan
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