Quiz sur les modèles de cohérence latente

Les réponses et les explications apparaissent au fur et à mesure que vous répondez au quiz.

1. What sampling goal does the cited LCM research investigate?
   • A guarantee of identical quality and latency on every phone
   • Useful image generation with substantially fewer sampling steps
   • Removal of all neural computation from generation
   • A mandatory increase to thousands of denoising steps
2. What consistency relationship applies to points on the same specified probability-flow trajectory?
   • Every unrelated image in the dataset must become identical
   • The function must ignore every input time
   • They map to the same near-clean endpoint
   • Each point must map to a different random filename
3. Dans quel espace les LCM effectuent-ils leur distillation ?
   • L'espace latent compressé d'un modèle de diffusion pré-entraîné
   • Espace de pixels bruts
   • Un espace d'intégration de texte séparé uniquement
   • Le domaine fréquentiel de l'audio
4. What does the LCM-LoRA report provide for its studied compatible Stable Diffusion families?
   • A guarantee that every arbitrary adapter matches every checkpoint
   • A replacement for all image decoding
   • A model trained from scratch by a single inference call
   • Low-rank acceleration parameters usable with compatible models and fine-tunes
5. What role does the pretrained guided diffusion model play in the original LCM distillation procedure?
   • It is merely an unrelated image filename list
   • It proves no training operation is needed
   • It supplies the teacher behavior used to train the consistency student
   • It replaces the associated autoencoder decoder entirely
6. Which cited related few-step method combines a teacher signal with an adversarial loss?
   • A JPEG compression setting
   • Adversarial Diffusion Distillation
   • An HTTP browser cache
   • A keyboard layout
7. What value should the consistency function f(x, epsilon) return at its stated near-clean boundary?
   • A required zero vector for every x
   • An unrelated random image identifier
   • The token count of a text prompt
   • x
8. What component turns the sampled latent representation into image pixels in the associated latent-diffusion setup?
   • The autoencoder decoder
   • Only the text tokenizer
   • Only a filename sorter
   • Only a data-split index