Temperature
Control randomness: Lowering results in less random completions. As the temperature approaches zero, the model will become deterministic and repetitive. Temperature is a parameter in OpenAI’s language model that controls the level of randomness or “creativity” in the generated text. The temperature parameter ranges from 0 to 1, with 0 being the most deterministic and 1 being the most unpredictable.
For example, if the temperature is set to 0, the model will generate text that is very similar to the training data and repetitive, while a temperature of 1 will result in more creative and varied text.
Max Tokens
Max Tokens is a parameter in OpenAI’s language model that controls the maximum number of tokens that can be generated in a single output. Tokens are the basic units of text, such as words or punctuation marks.
For example, if the max tokens is set to 50, the model will only generate a maximum of 50 tokens in its output. This can be used to limit the length of the generated text or to ensure that the model does not generate an excessive amount of text.
Top Prediction (Top-P)
Top Prediction (Top-P) is a parameter in OpenAI’s language model that controls the probability of the generated text being similar to the training data. This parameter is a way of controlling the level of randomness or “creativity” in the generated text.
For example, if the Top-P is set to 0.9, the model will generate text that is 90% similar to the training data, while a Top-P of 0.1 will result in more creative and varied text.
Best of
Best of is a parameter in OpenAI’s language model that controls the number of different outputs that will be generated for a single input. This parameter can be used to generate multiple variations of the same text, each with different levels of randomness or “creativity.”
For example, if the best of is set to 3, the model will generate three different outputs for a single input, each with a different level of randomness.
Frequency Penalty
Frequency Penalty is a parameter in OpenAI’s language model that controls the degree to which the model will avoid generating words or phrases that appear frequently in the training data. This parameter is a way of controlling the level of randomness or “creativity” in the generated text.
For example, if the frequency penalty is set to 0.5, the model will avoid generating words or phrases that frequently appear in the training data by 50%, while a frequency penalty of 1 will result in the model avoiding frequently used words and phrases entirely.
Presence Penalty
The Presence Penalty is a parameter in OpenAI’s language model that controls the degree to which the model will avoid generating words or phrases that were not present in the training data. This parameter is a way of controlling the level of randomness or “creativity” in the generated text.
For example, if the presence penalty is set to 0.5, the model will avoid generating words or phrases that were not present in the training data by 50%, while a presence penalty of 1 will result in the model avoiding all words or phrases that were not present in the training data.