Cross-Entropy Loss Explained: From Logits to Language Model Training
What is cross-entropy loss? Derive softmax + negative log-likelihood step by step, work a 3-class numeric example, see why MSE fails for classification, and build the loss GPT training actually uses.