passdrill
AI & LLM Engineering · How LLMs Work: Transformers & Training · Card 006/012 hard

Hoffmann et al. (2022), "Training Compute-Optimal Large Language Models" (the Chinchilla paper), trained over 400 models to study how model size and training data should be allocated for a fixed compute budget. What was their central finding?

  1. For a fixed training compute budget, model parameters and training tokens should be scaled at roughly the same rate, implying substantially larger training datasets relative to model size than many earlier large models had used
  2. Model performance becomes essentially independent of the number of training tokens once the parameter count exceeds a few billion
  3. Compute-optimal training requires increasing model size much faster than training data, since parameter count is the dominant driver of capability
  4. The optimal ratio of training tokens to parameters decreases as total training compute increases, so very large models need proportionally less data
Next card → Shuffle