Temperature

As explained in Build a Large Language Model (From Scratch)

A decoding method that rescales next-token probabilities with a temperature to make sampling more or less random.

How Build a Large Language Model (From Scratch) covers it

How deep it goes

Implements logit scaling, softmax conversion, and multinomial sampling, while deferring systematic setting combinations to exercises.

Examples

Demonstrates a 15-token continuation using temperature 1.4 and top-k 25, contrasting it with a memorized passage.

What it assumes

Assumes a model producing next-token logits and a tokenized prompt passed into the generation function.

What you can do after

Apply it via a modified generate function and compare outputs across temperature and top-k settings.

How it connects

Connects logits, softmax, multinomial sampling, top-k sampling, and autoregressive text generation.

liveBookmanning.com

Read Build a Large Language Model (From Scratch) on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Optimization Algorithms: AI techniques for design, planning, and control problemsA simulated-annealing control value that influences the likelihood of accepting worsening state transitions.
• 

  liveBookmanning.com
  Natural Language Processing in Action, Second EditionTemperature is a decoding parameter that rescales token scores to control the randomness of a generated distribution.
• 

  liveBookmanning.com
  Build a Reasoning Model (From Scratch)A token sampling method that draws from a probability distribution after temperature rescaling to control randomness.
• 

  liveBookmanning.com
  Build Python Web Apps with StreamlitTemperature is a language-model sampling parameter that controls how broadly probability is spread across possible next tokens.
• 

  liveBookmanning.com
  Generative AI in ActionTemperature is a model-generation setting that controls the degree of randomness in the model's output distribution.
• 

  liveBookmanning.com
  Data Analysis with LLMsTemperature is a model-generation setting that controls the degree of stochastic variation in selecting output tokens from their probability distribution.
• 

  liveBookmanning.com
  Effective Conversational AITemperature is a generation parameter that controls the degree of randomness in a language model's token selection.
• 

  liveBookmanning.com
  Building Reliable AI SystemsTemperature is a language-model generation parameter that controls the randomness of token selection.
• 

  liveBookmanning.com
  Deep Learning with R, Third EditionTemperature is a generation parameter that controls the concentration or spread of a model's next-token probability distribution.
• 

  liveBookmanning.com
  Machine Learning Platform EngineeringTemperature is a generation setting that adjusts how strongly token selection favors high-probability alternatives.
• 

  liveBookmanning.com
  AI Agents in Action, Second Edition
• 

  liveBookmanning.com
  Spring AI in ActionTemperature is a language-model generation parameter from 0 to 2 that scales token probabilities to control response randomness.
• 

  liveBookmanning.com
  Transformers in ActionTemperature sampling is a sampling strategy that adjusts the sharpness of a token probability distribution before selecting the next token.
• 

  liveBookmanning.com
  The Complete Obsolete Guide to Generative AITemperature is a generation parameter that controls the randomness or diversity of a language model's output.
• 

  liveBookmanning.com
  Build Applications with Local AI Models on a MacA language-model generation setting that controls the randomness of token selection.
• 

  liveBookmanning.com
  Machine Learning Algorithms in DepthA temperature parameter is a positive scaling value that controls the probability of accepting unfavorable transitions in simulated annealing.

Compare how 17 books explain temperature →

Related concepts

• softmaxuses · Softmax is a function that exponentiates a vector of scores and normalizes the results into nonnegative values that sum to one.
• logituses · A logit is a real-valued quantity representing an outcome’s log-odds or, in machine learning, an unnormalized model score used to derive probabilities.
• hyperparameterkind of · A hyperparameter is a setting chosen outside ordinary parameter learning that controls a model, algorithm, or learning procedure.
• output varianceconfigures · Output variance is the degree to which an AI model produces different results for ostensibly repeated or similar prompts.
• language modelconfigures · A language model is a computational model that represents patterns in linguistic tokens to estimate sequence probabilities and predict or generate language from context.

This explainer was generated by Manning's knowledge graph from the text of Build a Large Language Model (From Scratch). The section links point to the parts of the book it draws on.