AI tool

GradientBot AI

Step-by-step walkthroughs of training runs, loss curves and what the numbers mean.

Works in English · हिन्दी

Ask GradientBotHow GradientBot explains

How GradientBot explains

I am an AI tool, and I work in numbers. I take one small example from Machine Learning, Foundations or The Maths You Actually Need and show every line of the arithmetic: the forward pass, the loss, the gradient, the update. Give me your own values and I run the same step again, changing one thing at a time. In How a Language Model Actually Works and Fine-Tuning, and When Not To, I do the same with tokens and learning rates. When a question needs judgement rather than calculation, I say so and point you to an AI tutor.

Suits you if you want to see every number in a training step, with your own values plugged in.

A sample exchange

  1. My training loss printed nan after a few steps. What happened?

  2. GradientBot AI

    Take L(w) = (w - 3)², with slope 2(w - 3), starting at w = 0. With learning rate 0.1, w goes 0, 0.6, 1.08, 1.464 and closes in on 3. With 1.1, w goes 0, 6.6, -1.32, 8.18, -3.22, overshooting further each step until the numbers overflow and the loss prints nan. Lower the learning rate, usually by a factor of ten, and look again.1

    1. Machine Learning, Foundations · Lesson 2.3 · Gradient descent, or walking downhill in fog

A sample, grounded in that lesson. Your own question gets its own answer.

Courses it covers

Ask GradientBotSee all AI tutors