ManySpace
Türkçe Sign in Sign up

AI & MLRecommendedAIML-SYS-017

Mixed-Precision Training

Training in 16-bit or lower floating point with loss scaling to prevent underflow.

Difficulty
4 / 5
Importance
4 / 5
Study time
1–2 h
Kind of work
Mixed

Why it matters

The learner can compare fp32, fp16, and bf16 ranges and explain why loss scaling rescues small gradients in fp16.

The route to this topic

Route AIML-SYS-017
Natural Numbers What Is a Model? Variables & Constants Tasks: Regression, Classification, and Beyond Features and Targets Solving Linear Equations Function Notation & Composition Classification Framing Linear Regression I: Prediction and Residuals Complex Numbers Scores and Logits Loss and Fitting: Squared Error Basic Ratios Unit Circle The Artificial Neuron Objective Functions and Empirical Risk Polynomial Functions Activation Functions Intuitive Limits Continuity Definition of the Derivative Differentiation Rules The Multilayer Perceptron Chain Rule Antiderivatives & Indefinite Integrals Forward Propagation Derivatives of Transcendental Functions Riemann Sums & Definite Integral Training Dynamics Fundamental Theorem of Calculus Integration by Parts Floating Point Arithmetic Mixed-Precision Training