TinkerLoading step 0
Chapter 2 of 7
Watch it learn
Step 0 babbled. The run went on for 3,000 steps and saved 8 checkpoints along the way.
This chapter: How does the same prompt change as the model trains?
About this chapter
You will learn
- How each training step nudges the weights
- What the loss measures, and why it falls
- How a checkpoint lets you look back at any moment of a run
Words you’ll meet
- Training step
- One small change to the weights, made after the model reads a batch of text and checks its guesses.
- Loss
- How surprised the model is by the real next piece, on average. Lower is better; training pushes it down.
Step 1 of 4
The run made 3,000 . Each one nudged the weights so the next guess came out a little better.
Predict first.