One Answer, Four Different Scales
Interpret which parts of an LLM workflow change during training versus a single inference request.
A vendor PDF was uploaded for one procurement question. Which part of the workflow changed? Example Scales in One LLM Answer Training corpus Data used before serving to shape general language ability 10T tokens Model parameters Weights learned during training or fine-tuning 70B weights Current context window Prompt, retrieved PDF excerpts, conversation, and output budget for this request 32k tokens Generation step The live response is produced as a sequence of next-token choices 1 token at a time Widths are illustrative relative scales for separating training-time and inference-time concepts. When Lena uploads a vendor PDF for one question, which part…
Sign up free — one personalized lesson every day, matched to your role and goals.
Already have an account? Sign in