// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM
Temperature
A setting in AI models that controls how creative or random the generated output will be; higher temperatures lead to more varied and surprising text, while lower temperatures produce more focused and predictable text.
TECHNICAL DEFINITION
Temperature is a hyperparameter in large language models (LLMs) that controls the randomness of token prediction during text generation by scaling the logits before applying the softmax function, where higher values increase diversity and lower values promote determinism.
BACKGROUND
A large language model (LLM) is a neural network trained on a vast amount of text for natural language processing tasks, especially language generation. LLMs can typically generate, summarize, translate, and analyze text in many contexts, and are a foundational technology behind modern chatbots. Biased or inaccurate training data can make an LLM's output less reliable.
READ MORE ON WIKIPEDIASYNONYMS & ALIASES
- Creativity setting
- Randomness parameter
- Sampling temperature
- Output variability
USAGE NOTE
Adjusting temperature is a common technique for balancing creativity and coherence in LLM outputs.
DEVELOPERS
Organizations developing technology related to Temperature.
As the creator of the GPT series of models, OpenAI's API provides a 'temperature' parameter as a core setting for developers to control the randomness and creativity of text generation.
Through its Google AI and DeepMind divisions, Google develops models like Gemini. The Gemini API allows developers to adjust the temperature to fine-tune the model's output from deterministic to more creative.
Developer of the Claude family of AI models. Anthropic's API includes a temperature setting to modulate the randomness of the model's predictions, enabling users to balance creativity with coherence.
Provides the widely used 'transformers' library, which implements generation methods for thousands of models. The library gives developers direct control over sampling parameters like temperature to influence model output.
The developer of the Llama family of open-source models. Meta AI implements the core sampling algorithms, including temperature, that control the probabilistic nature of the model's token selection.
Cohere builds large language models for enterprise use cases. Its API for text generation features a temperature parameter, allowing businesses to adjust the level of invention in the model's responses.
An AI company developing both open-source and commercial language models. Their API and model configurations expose the temperature parameter to control the stochasticity of the output.
A cloud platform for training and running generative AI models. Its inference API provides standardized access to generation parameters, including temperature, across a wide variety of open-source models.