How deeply does LLMs in Production cover learning rate?

Explains parameter-update magnitude and gives a checkpoint-based adjustment procedure, stopping short of optimizer internals.

Learning rate

As explained in LLMs in Production

A learning rate is a training hyperparameter that controls the magnitude of parameter updates during optimization.

A learning rate controls how much model parameters change during optimization. A rate that is too large can prevent progress through difficult regions; a smaller rate can help a model move past an apparent but misleading convergence point. section 5.4.3 section 5.2.2

When training appears to stop before the dataset has been processed:

Treat early convergence as suspicious rather than accepting it immediately. section 5.4.3
Load a checkpoint from roughly 100 steps before the problematic behavior. section 5.4.3
Turn the learning rate down, train beyond the suspected local minimum, and raise it again after the model improves. section 5.4.3
Keep both the original and later checkpoints so failed recovery attempts remain reversible. section 5.4.3

Escaping repeated early convergence

A T5 XXL model appeared to converge around step 25,000 despite seeing only 10% of its dataset. Reloading a checkpoint near step 20,000 and lowering the learning rate let training pass the obstacle; the rate was raised afterward. section 5.4.3

How LLMs in Production covers it

liveBook manning.com

Read LLMs in Production on liveBook Get the book at manning.com

Also explained in

liveBook manning.com

Grokking Machine LearningA learning rate is a small positive training parameter that controls the size of parameter updates made during model optimization.
liveBook manning.com

Classic Computer Science Problems in Swift: Essential techniques for practicing programmersA learning rate is a training hyperparameter that controls the size of parameter updates made while fitting a model.
liveBook manning.com

Machine Learning for Tabular Data: XGBoost, Deep Learning, and AIA learning rate is a hyperparameter that scales the size of successive updates made while fitting a model.
liveBook manning.com

Inside Deep Learning: Math, Algorithms, ModelsThe step-size parameter controlling how far an optimization update moves in the direction suggested by the gradient.
liveBook manning.com

Ensemble Methods for Machine Learninglearning_rate is a boosting hyperparameter that scales the contribution of each successive weak learner and controls the rate at which ensemble complexity grows.
liveBook manning.com

Natural Language Processing in Action, Second EditionA learning rate is a numerical step-size setting that controls how far an optimization algorithm moves parameters in response to a gradient.
liveBook manning.com

Deep Learning with R, Third EditionA learning rate is the step-size hyperparameter controlling how far model parameters move during an optimization update.
liveBook manning.com

Deep Learning with PyTorch, Second EditionThe step-size hyperparameter that controls how much model parameters are changed per update during optimization.
liveBook manning.com

Machine Learning Platform EngineeringA learning rate is a training hyperparameter that controls the step size used to update model parameters during optimization.
liveBook manning.com

Math and Architectures of Deep LearningThe step-size setting that controls how far model parameters move on each optimization update, possibly chosen separately for each parameter dimension.
liveBook manning.com

Evolutionary Deep Learning: Genetic algorithms and neural networksA learning rate is the step-size parameter that determines how strongly model parameters are updated during training.
liveBook manning.com

Transformers in ActionLearning rate is an optimization hyperparameter that controls the magnitude of parameter updates during training.
liveBook manning.com

Data Without LabelsA learning rate is a scalar hyperparameter that controls the size of parameter updates during iterative training.
liveBook manning.com

Grokking AI Algorithms, Second EditionA learning rate is a training hyperparameter that controls the step size used to update model parameters.
liveBook manning.com

Designing Deep Learning Systems: A guide for software engineersA learning rate is a hyperparameter that controls the step size used to update model parameters during optimization.

Compare how 16 books explain learning rate →

Related concepts

hyperparameterkind of · A hyperparameter is a setting chosen outside ordinary parameter learning that controls a model, algorithm, or learning procedure.
gradient descentconfigures · Gradient descent is an iterative optimization method that updates parameters opposite the gradient of an objective function to reduce its value.
learning rate scheduleconfigured by · A learning rate schedule is a rule or function that determines how an optimizer’s learning rate changes during training.
stochastic gradient descentconfigures · Stochastic gradient descent is an iterative optimization method that updates parameters using gradients estimated from individual examples or randomly selected small batches.
cosine annealingconfigured by · Cosine annealing is a learning-rate scheduling method that varies the learning rate over training according to a cosine-shaped curve.

This explainer was generated by Manning's knowledge graph from the text of LLMs in Production. The section links point to the parts of the book it draws on.