Rajesh Srivastava on Instagram: "Part 2 - LLM Training Fundamentals Save this. Study this. You now know more than 90% of people talking about AI 1. BACKPROPAGATION - Gradients flow backward to update weights 2. CROSS-ENTROPY LOSS - Measures “surprise” at correct token 3. GRADIENT DESCENT - Walk downhill on the error landscape 4. LEARNING RATE - Step size (too big = explode, too small = slow) 5. OPTIMIZERS - AdamW is the LLM standard 6. LR SCHEDULE - Warmup → Peak → Cosine Decay 7. BATCH SIZE - L