Test your understanding of LLM fine-tuning, parameter alignment, and weight updating mechanics. This quiz covers SFT token loss masking, LoRA low-rank decomposition and matrix initialization, QLoRA NF4 quantization and memory optimizations, DPO implicit reward formulations, Elastic Weight Consolidation (EWC) Fisher matrix regularization, and KL divergence constraints against catastrophic forgetting.
0 / 6
In Supervised Fine-Tuning (SFT) on instruction-response pairs (prompt x, target response y), why is token-level loss masking applied to prompt tokens during cross-entropy loss computation?
Press 1 to 4 to pick an answer