Package com.microsoft.semantickernel.orchestration

Class PromptExecutionSettings

  • public class PromptExecutionSettings
extends Object
    Configuration settings for prompt execution.

    • Field Detail

      • DEFAULT_MAX_TOKENS

        public static final Integer DEFAULT_MAX_TOKENS

      • DEFAULT_TEMPERATURE

        public static final Double DEFAULT_TEMPERATURE

      • DEFAULT_TOP_P

        public static final Double DEFAULT_TOP_P

      • DEFAULT_PRESENCE_PENALTY

        public static final Double DEFAULT_PRESENCE_PENALTY

      • DEFAULT_FREQUENCY_PENALTY

        public static final Double DEFAULT_FREQUENCY_PENALTY

      • DEFAULT_BEST_OF

        public static final Integer DEFAULT_BEST_OF

      • DEFAULT_RESULTS_PER_PROMPT

        public static final Integer DEFAULT_RESULTS_PER_PROMPT

    • Constructor Detail

      • PromptExecutionSettings

        public PromptExecutionSettings​(String serviceId,
                               String modelId,
                               Double temperature,
                               Double topP,
                               Double presencePenalty,
                               Double frequencyPenalty,
                               Integer maxTokens,
                               Integer resultsPerPrompt,
                               Integer bestOf,
                               String user,
                               @Nullable
                               List<String> stopSequences,
                               @Nullable
                               Map<Integer,​Integer> tokenSelectionBiases)
        Create a new instance of PromptExecutionSettings.
        Parameters:
        serviceId - The id of the AI service to use for prompt execution.
        modelId - The id of the model to use for prompt execution.
        temperature - The temperature setting for prompt execution.
        topP - The topP setting for prompt execution.
        presencePenalty - The presence penalty setting for prompt execution.
        frequencyPenalty - The frequency penalty setting for prompt execution.
        maxTokens - The maximum number of tokens to generate in the output.
        resultsPerPrompt - The number of results to generate for each prompt.
        bestOf - The best of setting for prompt execution.
        user - The user to associate with the prompt execution.
        stopSequences - The stop sequences to use for prompt execution.
        tokenSelectionBiases - The token selection biases to use for prompt execution.

    • Method Detail

      • builder

        public static PromptExecutionSettings.Builder builder()
        Create a new builder for PromptExecutionSettings.
        Returns:
        A new builder for PromptExecutionSettings.

      • getServiceId

        public String getServiceId()
        Get the id of the AI service to use for prompt execution.
        Returns:
        The id of the AI service to use for prompt execution.

      • getModelId

        public String getModelId()
        Get the id of the model to use for prompt execution.
        Returns:
        The id of the model to use for prompt execution.

      • getTemperature

        public double getTemperature()
        The temperature setting controls the randomness of the output. Lower values produce more deterministic outputs, while higher values produce more random outputs.
        Returns:
        The temperature setting.

      • getTopP

        public double getTopP()
        The topP setting controls how many different words or phrases are considered to predict the next token. The value is a probability threshold, and the model considers the most likely tokens whose cumulative probability mass is greater than the threshold. For example, if the value is 0.1, the model considers only the tokens that make up the top 10% of the cumulative probability mass.
        Returns:
        The topP setting.

      • getPresencePenalty

        public double getPresencePenalty()
        Presence penalty encourages the model to use a more or less diverse range of tokens in the output. A higher value means that the model will try to use a greater variety of tokens in the ouput.
        Returns:
        The presence penalty setting.

      • getFrequencyPenalty

        public double getFrequencyPenalty()
        Frequency penalty encourages the model to avoid repeating the same token in the output. A higher value means that the model will be less likely to repeat a token.
        Returns:
        The frequency penalty setting.

      • getMaxTokens

        public int getMaxTokens()
        The maximum number of tokens to generate in the output.
        Returns:
        The maximum number of tokens to generate in the output.

      • getResultsPerPrompt

        public int getResultsPerPrompt()
        The number of results to generate for each prompt.
        Returns:
        The number of results to generate for each prompt.

      • getBestOf

        public int getBestOf()
        The log probability threshold for a result to be considered.
        Returns:
        The log probability threshold for a result to be considered.

      • getUser

        public String getUser()
        The user to associate with the prompt execution.
        Returns:
        The user to associate with the prompt execution.

      • getStopSequences

        @Nullable
public List<String> getStopSequences()
        The stop sequences to use for prompt execution.
        Returns:
        The stop sequences to use for prompt execution.

      • getTokenSelectionBiases

        @Nullable
public Map<Integer,​Integer> getTokenSelectionBiases()
        The token selection biases to use for prompt execution. The key is the token id from the tokenizer, and the value is the bias. A negative bias will make the model less likely to use the token, and a positive bias will make the model more likely to use the token.
        Returns:
        The token selection biases to use for prompt execution.