PaLMr - Documentation

Last update: 2024/01/21

'Google Pathways Language Model 2 (PaLM 2)' as a coding and writing assistant designed for 'R'. With a range of functions, including natural language processing and coding optimization, to assist 'R' developers in simplifying tedious coding tasks and content searching.

⚠️NOTICE⚠️:
As Google Gemini is on its way, PaLM becomes legacy. We recommend that you use the latest R package, GenAI: Generative Artificial Intelligence. This package allows you to seamlessly utilize both ChatGPT from OpenAI and Gemini Pro from Google!

Prerequisites

Prior to utilizing the PaLMr package, several prerequisites must be met.

  1. Ensure that you possess an eligible device equipped with R.

  2. Access to the internet is essential to generate text or engage in chat through PaLMr.

  3. Obtain an API key from Google. To acquire an API key for Google's models, refer to: Get an API key

Installation

1. Install the package from The Comprehensive R Archive Network (CRAN). 

install.packages("PaLMr")

2. Install the package from PaLMr official website. 

install.packages("https://palmr.ly.gd.edu.kg/release/PaLMr_latest.tar.gz", repos=NULL, method="libcurl")

Setup

palm.connect
Open in GitHub

Set up and connect to the Google PaLM 2 text model

Usage
palm.connect(version, api, proxy = FALSE)
Value

If successful, the function returns a character vector containing the API key, model version, and proxy status. If the API response indicates an error, the function stops execution and provides an error message.

Description

This function establishes a connection to the Google PaLM text model by specifying the API key, and model version.

Detail

This function performs the necessary setup to connect to the Google PaLM 2 text model. It validates the provided API key and checks the correctness of the model version. If the input is valid, it constructs the API request and sends it to the PaLM 2 API endpoint.

If an error occurs during the API request, such as an invalid API key or input parameters, an error message is displayed. If the API request is successful, the function prints the model details to the console and returns a character vector with the API key, model version, and model type.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers

Text Generation

palm.txt
Open in GitHub

Generate text using the Google PaLM 2 text model based on a prompt

Usage
palm.txt(model.parameter,
         prompt,
         temperature = 0.7,
         maxOutputTokens = 1024,
         topP = 0.95,
         topK = 40,
         htUnspecified = "meda",
         htDerogatory = "meda",
         htToxicity = "meda",
         htViolence = "meda",
         htSexual = "meda",
         htMedical = "meda",
         htDangerous = "meda")
Arguments
  • model.parameter: A character vector containing the API key, model version, and proxy status. Model version and type are specified by Google. See function palm.connect for detail.
  • prompt: A character string representing the query or prompt for text generation. The length of the query should be between 1 and 8196 characters, inclusive.
  • temperature: A numeric value between 0.0 and 1.0, inclusive (default: 0.7). Controls the randomness of the generated text. A higher value (e.g., 0.9) results in more creative responses, while a lower value (e.g., 0.3) produces more straightforward text.
  • maxOutputTokens: An integer value (default: 1024). Specifies the maximum number of tokens to include in the generated text.
  • topP: A numeric value (default: 0.95). Defines the maximum cumulative probability of tokens considered when sampling. It controls the diversity of the text generated.
  • topK: An integer value (default: 40). Sets the maximum number of tokens to consider when sampling.
  • htUnspecified: Safety setting threshold for unspecified harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDerogatory: Safety setting threshold for derogatory harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htToxicity: Safety setting threshold for toxicity harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htViolence: Safety setting threshold for violence harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htSexual: Safety setting threshold for sexual harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htMedical: Safety setting threshold for medical harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDangerous: Safety setting threshold for dangerous harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
Value

A character string generated by the Google PaLM 2 API based on the provided prompt and parameters.

Description

This function sends a prompt to the Google PaLM 2 text model and generates text as a response. It allows customization of the generated text using various parameters.

Detail

This function interacts with the Google PaLM model by sending a query using the specified parameters. It allows you to customize the generated text by adjusting the temperature, maxOutputTokens, topP, topK, and safety settings.

If the function is successful, it returns a character string containing the generated text. If an error occurs during the API request, it will stop execution and provide an error message.

The model.parameter argument should be a character vector with the API key, model version, and model type provided by Google. You can obtain this information by following the instructions provided by Google for using the PaLM API.

The safety settings control the content's safety level based on different harm categories. Harm thresholds are specified as per Google's guidelines and can be customized to control the content generated.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model

prompt = "Write a story about a magic backpack."
generated.text = palm.txt(palm.model,
                          prompt)
cat(generated.text)
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers

palm.txt.explain.code
Open in GitHub

Explain code based on a query using the Google PaLM 2 text model

Usage
palm.txt.explain.code(model.parameter,
                      prompt,
                      language = "R",
                      temperature = 0.7,
                      maxOutputTokens = 1024,
                      topP = 0.95,
                      topK = 40,
                      htUnspecified = "meda",
                      htDerogatory = "meda",
                      htToxicity = "meda",
                      htViolence = "meda",
                      htSexual = "meda",
                      htMedical = "meda",
                      htDangerous = "meda")
Arguments
  • model.parameter: A character vector containing the API key, model version, and proxy status. Model version and type are specified by Google. See function palm.connect for detail.
  • prompt: A character string representing the query or prompt for text generation. The length of the query should be between 1 and 8196 characters, inclusive.
  • language: A character string specifying the programming language used in the code (default: "R").
  • temperature: A numeric value between 0.0 and 1.0, inclusive (default: 0.7). Controls the randomness of the generated text. A higher value (e.g., 0.9) results in more creative responses, while a lower value (e.g., 0.3) produces more straightforward text.
  • maxOutputTokens: An integer value (default: 1024). Specifies the maximum number of tokens to include in the generated text.
  • topP: A numeric value (default: 0.95). Defines the maximum cumulative probability of tokens considered when sampling. It controls the diversity of the text generated.
  • topK: An integer value (default: 40). Sets the maximum number of tokens to consider when sampling.
  • htUnspecified: Safety setting threshold for unspecified harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDerogatory: Safety setting threshold for derogatory harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htToxicity: Safety setting threshold for toxicity harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htViolence: Safety setting threshold for violence harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htSexual: Safety setting threshold for sexual harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htMedical: Safety setting threshold for medical harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDangerous: Safety setting threshold for dangerous harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE

A character string containing the detailed explanation of the provided code snippet based on the query and parameters.

Description

This function sends a query with a code snippet to the Google PaLM 2 text model and generates a detailed explanation of the code. It supports various programming languages and allows you to customize the explanation.

Detail

This function interacts with the Google PaLM model by sending a code query to explain code. It allows you to customize the generated code explanations by specifying the programming language, and additional parameters like temperature, token limits, and safety settings.

If the function is successful, it returns a detailed explanation of the provided code as a character string. If an error occurs during the API request, it will stop execution and provide an error message.

The model.parameter argument should be a character vector with the API key, model version, and model type provided by Google. You can obtain this information by following the instructions provided by Google for using the PaLM API.

The safety settings control the content's safety level based on different harm categories. Harm thresholds are specified as per Google's guidelines and can be customized to control the content generated.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model

prompt = "foo <- function(n) {
            if (n <= 0) {
              return(0)
            } else if (n == 1) {
              return(1)
            } else {
              return(foo(n - 1) + foo(n - 2))
            }
          }"
code.explanation = palm.txt.explain.code(palm.model,
                                         prompt)
cat(code.explanation)
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers

palm.txt.fix.grammar
Open in GitHub

Fix grammar and rewrite text using the Google PaLM 2 text model

Usage
palm.txt.fix.grammar(model.parameter,
                     prompt,
                     temperature = 0.7,
                     maxOutputTokens = 1024,
                     topP = 0.95,
                     topK = 40,
                     htUnspecified = "meda",
                     htDerogatory = "meda",
                     htToxicity = "meda",
                     htViolence = "meda",
                     htSexual = "meda",
                     htMedical = "meda",
                     htDangerous = "meda")
Arguments
  • model.parameter: A character vector containing the API key, model version, and proxy status. Model version and type are specified by Google. See function palm.connect for detail.
  • prompt: A character string representing the query or prompt for text generation. The length of the query should be between 1 and 8196 characters, inclusive.
  • language: A character string specifying the programming language used in the code (default: "R").
  • temperature: A numeric value between 0.0 and 1.0, inclusive (default: 0.7). Controls the randomness of the generated text. A higher value (e.g., 0.9) results in more creative responses, while a lower value (e.g., 0.3) produces more straightforward text.
  • maxOutputTokens: An integer value (default: 1024). Specifies the maximum number of tokens to include in the generated text.
  • topP: A numeric value (default: 0.95). Defines the maximum cumulative probability of tokens considered when sampling. It controls the diversity of the text generated.
  • topK: An integer value (default: 40). Sets the maximum number of tokens to consider when sampling.
  • htUnspecified: Safety setting threshold for unspecified harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDerogatory: Safety setting threshold for derogatory harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htToxicity: Safety setting threshold for toxicity harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htViolence: Safety setting threshold for violence harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htSexual: Safety setting threshold for sexual harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htMedical: Safety setting threshold for medical harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDangerous: Safety setting threshold for dangerous harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
Value

A character string containing the rewritten text with corrected grammar, generated by the Google PaLM API based on the provided query and parameters.

Description

This function sends a query with grammatical issues to the Google PaLM 2 text model and generates corrected text as a response. It allows customization of the generated text using various parameters.

Detail

This function interacts with the Google PaLM model by sending a query with grammatical issues using the specified parameters. It allows you to customize the generated text by adjusting the temperature, maxOutputTokens, topP, topK, and safety settings.

If the function is successful, it returns a character string containing the rewritten text with corrected grammar. If an error occurs during the API request, it will stop execution and provide an error message.

The model.parameter argument should be a character vector with the API key, model version, and model type provided by Google. You can obtain this information by following the instructions provided by Google for using the PaLM API.

The safety settings control the content's safety level based on different harm categories. Harm thresholds are specified as per Google's guidelines and can be customized to control the content generated.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model

prompt = "Yesterday, I will buy a book for my younger sister as his birthday gift.
          They were very happen when seeing this gift earlier today."
correct.text = palm.txt.fix.grammar(palm.model,
                                    prompt)
cat(correct.text)
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers

palm.txt.get.reference
Open in GitHub

Get references based on a query using the Google PaLM 2 text model

Usage
palm.txt.get.reference(model.parameter,
                       prompt,
                       source.type = "articles",
                       source.date = "most recent",
                       n.source = 5,
                       citation.style = "APA7",
                       temperature = 0.7,
                       maxOutputTokens = 1024,
                       topP = 0.95,
                       topK = 40,
                       htUnspecified = "meda",
                       htDerogatory = "meda",
                       htToxicity = "meda",
                       htViolence = "meda",
                       htSexual = "meda",
                       htMedical = "meda",
                       htDangerous = "meda")
Arguments
  • model.parameter: A character vector containing the API key, model version, and proxy status. Model version and type are specified by Google. See function palm.connect for detail.
  • prompt: A character string representing the query or prompt for text generation. The length of the query should be between 1 and 8196 characters, inclusive.
  • source.type: A character string specifying the type of sources to search for (default: "articles").
  • source.date: A character string specifying the date range for the sources (default: "most recent").
  • n.source: An integer value specifying the number of sources to retrieve (default: 5).
  • citation.style: A character string specifying the citation style for the references (default: "APA7").
  • temperature: A numeric value between 0.0 and 1.0, inclusive (default: 0.7). Controls the randomness of the generated text. A higher value (e.g., 0.9) results in more creative responses, while a lower value (e.g., 0.3) produces more straightforward text.
  • maxOutputTokens: An integer value (default: 1024). Specifies the maximum number of tokens to include in the generated text.
  • topP: A numeric value (default: 0.95). Defines the maximum cumulative probability of tokens considered when sampling. It controls the diversity of the text generated.
  • topK: An integer value (default: 40). Sets the maximum number of tokens to consider when sampling.
  • htUnspecified: Safety setting threshold for unspecified harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDerogatory: Safety setting threshold for derogatory harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htToxicity: Safety setting threshold for toxicity harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htViolence: Safety setting threshold for violence harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htSexual: Safety setting threshold for sexual harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htMedical: Safety setting threshold for medical harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDangerous: Safety setting threshold for dangerous harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
Value

A character string containing the generated references based on the provided query and parameters.

Description

This function sends a query to the Google PaLM 2 text model and generates a list of references based on the query. It allows customization of the generated references and supports various citation styles and source types.

Detail

This function interacts with the Google PaLM model by sending a query to find references. It allows you to customize the generated references by specifying the number of sources, citation style, source type, date range, and safety settings.

If the function is successful, it returns a character vector containing the generated references. If an error occurs during the API request, it will stop execution and provide an error message.

The model.parameter argument should be a character vector with the API key, model version, and model type provided by Google. You can obtain this information by following the instructions provided by Google for using the PaLM API.

The safety settings control the content's safety level based on different harm categories. Harm thresholds are specified as per Google's guidelines and can be customized to control the content generated.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model

correct.text = palm.txt.get.reference(palm.model,
                                      "H5N1 in the United States")
cat(correct.text)
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers

palm.txt.optimize.code
Open in GitHub

Optimize code based on a query using the Google PaLM 2 text model

Usage
palm.txt.optimize.code(model.parameter,
                       prompt,
                       goal,
                       language = "R",
                       temperature = 0.7,
                       maxOutputTokens = 1024,
                       topP = 0.95,
                       topK = 40,
                       htUnspecified = "meda",
                       htDerogatory = "meda",
                       htToxicity = "meda",
                       htViolence = "meda",
                       htSexual = "meda",
                       htMedical = "meda",
                       htDangerous = "meda")
Arguments
  • model.parameter: A character vector containing the API key, model version, and proxy status. Model version and type are specified by Google. See function palm.connect for detail.
  • prompt: A character string representing the query or prompt for text generation. The length of the query should be between 1 and 8196 characters, inclusive.
  • goal: A character string specifying the aspect you want to optimize.
  • language: A character string specifying the programming language used in the code (default: "R").
  • temperature: A numeric value between 0.0 and 1.0, inclusive (default: 0.7). Controls the randomness of the generated text. A higher value (e.g., 0.9) results in more creative responses, while a lower value (e.g., 0.3) produces more straightforward text.
  • maxOutputTokens: An integer value (default: 1024). Specifies the maximum number of tokens to include in the generated text.
  • topP: A numeric value (default: 0.95). Defines the maximum cumulative probability of tokens considered when sampling. It controls the diversity of the text generated.
  • topK: An integer value (default: 40). Sets the maximum number of tokens to consider when sampling.
  • htUnspecified: Safety setting threshold for unspecified harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDerogatory: Safety setting threshold for derogatory harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htToxicity: Safety setting threshold for toxicity harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htViolence: Safety setting threshold for violence harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htSexual: Safety setting threshold for sexual harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htMedical: Safety setting threshold for medical harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
  • htDangerous: Safety setting threshold for dangerous harm. The default threshold is "meda". Valid options are as follows.
    • "unsp": HARM_BLOCK_THRESHOLD_UNSPECIFIED
    • "lowa": BLOCK_LOW_AND_ABOVE
    • "meda": BLOCK_MEDIUM_AND_ABOVE
    • "high": BLOCK_ONLY_HIGH
    • "none": BLOCK_NONE
Value

A character string containing the optimized version of the provided code snippet based on the query and parameters.

Description

This function sends a query with a code snippet to the Google PaLM 2 text model and generates an optimized version of the code. You can specify the programming language and the aspect you want to optimize (e.g., "runtime" or "memory"). The optimized code is provided along with the original code for comparison.

Detail

This function interacts with the Google PaLM model by sending a code query for code optimization. It allows you to customize the generated code optimizations by specifying the programming language, optimization aspect, and additional parameters like temperature, token limits, and safety settings.

If the function is successful, it returns an optimized version of the provided code as a character string. If an error occurs during the API request, it will stop execution and provide an error message.

The model.parameter argument should be a character vector with the API key, model version, and model type provided by Google. You can obtain this information by following the instructions provided by Google for using the PaLM API.

The safety settings control the content's safety level based on different harm categories. Harm thresholds are specified as per Google's guidelines and can be customized to control the content generated.

Open In Colab 
palm.model = palm.connect("v1beta3",
                          "API_KEY",
                          FALSE)
palm.model

prompt = "foo <- function(n) {
            if (n <= 0) {
              return(0)
            } else if (n == 1) {
              return(1)
            } else {
              return(foo(n - 1) + foo(n - 2))
            }
          }"
code.optimization = palm.txt.optimize.code(palm.model,
                                           "Improve the runtime.",
                                           prompt)
cat(code.optimization)
See Also

Safety Setting - Google AI for Developers

HarmCategory - Google AI for Developers