--- title: "LLMFunction" language: "en" type: "Symbol" summary: "LLMFunction[prompt] represents a template for a large language model (LLM) prompt. LLMFunction[{prompt1, prompt2, ...}] represents a combination of multiple prompts. LLMFunction[prompt, form] includes the interpreter form to apply to the response. LLMFunction[...][params] give the LLM service response for prompt applied to parameters params." keywords: - Autocomplete - Computed responses - Context - Contextual data handling - Data-based prompt responses - Data-driven text generation - Deep learning - Dynamic content creation - External tools - GPT - Language model - Language models - Language processing - LLM - Machine learning - Machine translation - Natural Language Processing (NLP) - Natural language understanding - Neural network - Prediction - Prompt - Real-time data integration - Semantic processing - Sentence generation - Sequence generation - Text analysis - Text generation - Text prediction - Text synthesis - Tool interfacing canonical_url: "https://reference.wolfram.com/language/ref/LLMFunction.html" source: "Wolfram Language Documentation" related_guides: - title: "Machine Learning" link: "https://reference.wolfram.com/language/guide/MachineLearning.en.md" - title: "Text Manipulation" link: "https://reference.wolfram.com/language/guide/ProcessingTextualData.en.md" - title: "LLM-Related Functionality" link: "https://reference.wolfram.com/language/guide/LLMFunctions.en.md" - title: "Text Analysis" link: "https://reference.wolfram.com/language/guide/TextAnalysis.en.md" - title: "Natural Language Processing" link: "https://reference.wolfram.com/language/guide/NaturalLanguageProcessing.en.md" - title: "String Manipulation" link: "https://reference.wolfram.com/language/guide/StringManipulation.en.md" - title: "Working with Templates" link: "https://reference.wolfram.com/language/guide/WorkingWithTemplates.en.md" - title: "Converting between Expressions & Strings" link: "https://reference.wolfram.com/language/guide/ConvertingBetweenExpressionsAndStrings.en.md" - title: "Linguistic Data" link: "https://reference.wolfram.com/language/guide/LinguisticData.en.md" - title: "Free-Form & External Input" link: "https://reference.wolfram.com/language/guide/FreeFormAndExternalInput.en.md" - title: "Setting Up Input Interpreters" link: "https://reference.wolfram.com/language/guide/InterpretingStrings.en.md" - title: "Knowledge Representation & Access" link: "https://reference.wolfram.com/language/guide/KnowledgeRepresentationAndAccess.en.md" - title: "Text Generation" link: "https://reference.wolfram.com/language/guide/TextConstruction.en.md" - title: "Text Normalization" link: "https://reference.wolfram.com/language/guide/TextNormalization.en.md" --- [EXPERIMENTAL] # LLMFunction [image] This functionality requires [LLM access](https://www.wolfram.com/notebook-assistant-llm-kit/) LLMFunction[prompt] represents a template for a large language model (LLM) prompt. LLMFunction[{prompt1, prompt2, …}] represents a combination of multiple prompts. LLMFunction[prompt, form] includes the interpreter form to apply to the response. LLMFunction[…][params] give the LLM service response for prompt applied to parameters params. ## Details and Options * An ``LLMFunction`` can be used to generate text using a large language model (LLM). It can create content, complete sentences, extract information and more. * ``LLMFunction`` requires external service authentication, billing and internet connectivity. * The ``prompti`` supports the following values: | | | | ----------------- | ------------------------------ | | "string" | static string template | | LLMPrompt["name"] | a repository prompt | | StringTemplate[…] | templated text | | TemplateObject[…] | template for creating a prompt | | {prompt1, …} | a list of prompts | * A prompt ``"string"`` is equivalent to ``StringTemplate["string"]``. * Prompts created with ``TemplateObject`` can contain text and images. Not every LLM supports image input. * The supported values of ``form`` are the same as ``Interpreter``. * When a ``form`` is specified, the LLM will be instructed to respect it. * Specifying an output interpretation adds the corresponding schema information to the LLM prompt. * ``LLMFunction`` supports the following options: | | | | | ----------------- | -------------- | -------------------------------------------------------- | | InsertionFunction | TextString | function or format to apply before inserting expressions | | CombinerFunction | StringJoin | function to apply to combine pieces within a prompt | | Authentication | Automatic | explicit user ID and API key | | LLMEvaluator | \$LLMEvaluator | LLM configuration to use | * ``LLMEvaluator`` can be set to an ``LLMConfiguration`` object or an association with any of the following keys: | | | | ------------------------ | ---------------------------------------------- | | "MaxTokens" | maximum amount of tokens to generate | | "Model" | base model | | "PromptDelimiter" | string to insert between prompts | | "Prompts" | initial prompts | | "StopTokens" | tokens on which to stop generation | | "Temperature" | sampling temperature | | "ToolMethod" | method to use for tool calling | | "Tools" | list of LLMTool objects to make available | | "TopProbabilities" | sampling classes cutoff | | "TotalProbabilityCutoff" | sampling probability cutoff (nucleus sampling) | * Valid forms of ``"Model"`` include: | | | | ------------------------------------------------------- | ------------------------ | | name | named model | | {service, name} | named model from service | | <\|"Service" -> service, "Name" -> name, "Task" -> task\|> | fully specified model | * The generated text is sampled from a distribution. Details of the sampling can be specified using the following properties of ``LLMEvaluator`` : | | | | | --- | --- | --- | | "Temperature" -> t | Automatic | sample using a positive temperature t | | "TopProbabilities" -> k | Automatic | sample only among the k highest-probability classes | | "TotalProbabilityCutoff" -> p | Automatic | sample among the most probable choices with an accumulated probability of at least p (nucleus sampling) | * Multiple prompts are separated by the ``"PromptDelimiter"`` property of the ``LLMEvaluator``. * Possible values for ``Authentication`` are: | | | | ---------------- | ------------------------------------------------ | | Automatic | choose the authentication scheme automatically | | Environment | check for a key in the environment variables | | SystemCredential | check for a key in the system keychain | | ServiceObject[…] | inherit the authentication from a service object | | assoc | provide an explicit key and user ID | * With ``Authentication -> Automatic``, the function checks the variable ``ToUpperCase[service] <> "_API_KEY"`` in ``Environment`` and ``SystemCredential``; otherwise, it uses ``ServiceConnect[service]``. * When using ``Authentication -> assoc``, ``assoc`` can contain the following keys: | | | | -------- | ---------------------------- | | "ID" | user identity | | "APIKey" | API key used to authenticate | * ``LLMFunction`` uses machine learning. Its methods, training sets and biases included therein may change and yield varied results in different versions of the Wolfram Language. ## Examples (8) ### Basic Examples (3) Create a function for getting cooking instructions: ```wl In[1]:= LLMFunction["Describe how to cook `1`"] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "Anthropic", "Name" -> "claude-3-5-haiku-20241022", "ServiceObject" -> ServiceObject["Anthropic", Association["ID" -> ... }, "Template" -> TemplateObject[{"Describe how to cook ", TemplateSlot[1]}, CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> "String", "LLMPacletVersion" -> "2.2.0"]] ``` --- Create a helper tool: ```wl In[1]:= findbyproperty = LLMFunction["What is something that is `1` and `2`"] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "OpenAI", "Name" -> "gpt-4o-mini", "ServiceObject" -> ServiceObject["OpenAI", Association[ "ID" -> "yXqt3eiO8bodRCCF3z ... ct[{"What is something that is ", TemplateSlot[1], " and ", TemplateSlot[2]}, CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> "String", "LLMPacletVersion" -> "2.2.0"]] ``` Use the function: ```wl In[2]:= findbyproperty["Blue", "Round"] Out[2]= "One example of something that is blue and round is a blue marble. Other examples could include a blue balloon, a blue beach ball, or a blue planet like Earth when viewed from space." ``` --- Create a function that returns a city as an ``Entity`` : ```wl In[1]:= cityfinder = LLMFunction["Find a city between `1` and `2`. Answer with the city name only", "City"] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "OpenAI", "Name" -> "gpt-4o-mini", "ServiceObject" -> ServiceObject["OpenAI", Association[ "ID" -> "yXqt3eiO8bodRCCF3z ... plateSlot[1], " and ", TemplateSlot[2], ". Answer with the city name only"}, CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> "City", "LLMPacletVersion" -> "2.2.0"]] ``` Use the function: ```wl In[2]:= cityfinder["St. Louis", "Chicago"] Out[2]= Entity["City", {"Springfield", "Illinois", "UnitedStates"}] ``` ### Scope (4) Use named parameters: ```wl In[1]:= teamfinder = LLMFunction["What team did `Pitcher` and `Catcher` play for?"] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "OpenAI", "Name" -> "gpt-4o-mini", "ServiceObject" -> ServiceObject["OpenAI", Association[ "ID" -> "yXqt3eiO8bodRCCF3z ... TemplateSlot["Pitcher"], " and ", TemplateSlot["Catcher"], " play for?"}, CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> "String", "LLMPacletVersion" -> "2.2.0"]] ``` Apply it using an ``Association`` : ```wl In[2]:= teamfinder[<|"Pitcher" -> "Whitey Ford", "Catcher" -> "Yogi Berra"|>] Out[2]= "Whitey Ford and Yogi Berra both played for the New York Yankees. They were key figures in the team's success during the 1950s and 1960s." ``` --- Set a default for a parameter: ```wl In[1]:= recipe = LLMFunction[StringTemplate["Create a short recipe including `Veggie` annd `Spice`", <|"Spice" -> "Chili Powder"|>]] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "OpenAI", "Name" -> "gpt-4o-mini", "ServiceObject" -> ServiceObject["OpenAI", Association[ "ID" -> "yXqt3eiO8bodRCCF3z ... " annd ", TemplateSlot["Spice"]}, Association["Spice" -> "Chili Powder"], CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> "String", "LLMPacletVersion" -> "2.2.0"]] In[2]:= recipe[<|"Veggie" -> "Corn"|>] Out[2]= "### Spicy Corn Salad #### Ingredients: - 2 cups fresh or frozen corn (thawed if frozen) - 1 tablespoon olive oil - 1 teaspoon chili powder - 1/2 teaspoon salt - 1/4 teaspoon black pepper - 1/2 cup diced red bell pepper - 1/4 cup chopped fresh cil ... ntil the corn and peppers are evenly coated with the dressing. 4. **Serve**: Let the salad sit for about 10 minutes to allow the flavors to meld. Serve chilled or at room temperature. Enjoy your spicy corn salad as a side dish or a light snack!" ``` --- Use a multi-part prompt with images: ```wl In[1]:= LLMFunction[TemplateObject@{"what is this?", TemplateSlot[1]}][[image]] Out[1]= "This is a slice of lemon. Lemons are citrus fruits known for their bright yellow color and tart flavor. They are commonly used in cooking, baking, and beverages." ``` --- Use an ``Interpreter`` to both constrain and validate the result: ```wl In[1]:= f = LLMFunction["Generate `` random instances", RepeatingElement["Color"]] Out[1]= LLMFunction[Association["LLMEvaluator" -> LLMConfiguration[ Association["Model" -> Association["Service" -> "LLMKitAzureOpenAI", "Name" -> Automatic, "Alias" -> "Wolfram LLM Service", "ServiceObject" -> ServiceObject["LLMKitAzureOpenAI" ... bject[{"Generate ", TemplateSlot[1], " random instances"}, CombinerFunction -> StringJoin, InsertionFunction -> TextString, MetaInformation -> Association[]], "Interpreter" -> RepeatingElement["Color"], "LLMPacletVersion" -> "2.2.0"]] In[2]:= f[10] Out[2]= {RGBColor[1, 0, 0], RGBColor[0, 0, 1], RGBColor[0, 1, 0], RGBColor[1, 1, 0], RGBColor[0.5, 0, 0.5], RGBColor[1, 0.5, 0], RGBColor[1, 0.5, 0.5], RGBColor[0.6, 0.4, 0.2], RGBColor[0, 1, 1], RGBColor[1, 0, 1]} ``` ### Properties & Relations (1) ``LLMFunction`` with no parameters sends the prompt directly to the LLM: ```wl In[1]:= LLMFunction["Hello"][] Out[1]= "Hello! How can I assist you today?" ``` This is equivalent to ``LLMSynthesize`` with zero temperature: ```wl In[2]:= LLMSynthesize["Hello", LLMEvaluator -> <|"Temperature" -> 0|>] Out[2]= "Hello! How can I assist you today?" ``` ## See Also * [`LLMExampleFunction`](https://reference.wolfram.com/language/ref/LLMExampleFunction.en.md) * [`LLMSynthesize`](https://reference.wolfram.com/language/ref/LLMSynthesize.en.md) * [`LLMGraph`](https://reference.wolfram.com/language/ref/LLMGraph.en.md) * [`LLMPrompt`](https://reference.wolfram.com/language/ref/LLMPrompt.en.md) * [`LLMTool`](https://reference.wolfram.com/language/ref/LLMTool.en.md) * [`OpenAI`](https://reference.wolfram.com/language/ref/service/OpenAI.en.md) * [`Anthropic`](https://reference.wolfram.com/language/ref/service/Anthropic.en.md) * [`GoogleGemini`](https://reference.wolfram.com/language/ref/service/GoogleGemini.en.md) * [`AlephAlpha`](https://reference.wolfram.com/language/ref/service/AlephAlpha.en.md) * [`Cohere`](https://reference.wolfram.com/language/ref/service/Cohere.en.md) * [`DeepSeek`](https://reference.wolfram.com/language/ref/service/DeepSeek.en.md) * [`Groq`](https://reference.wolfram.com/language/ref/service/Groq.en.md) * [`MistralAI`](https://reference.wolfram.com/language/ref/service/MistralAI.en.md) * [`TogetherAI`](https://reference.wolfram.com/language/ref/service/TogetherAI.en.md) ## Related Guides * [Machine Learning](https://reference.wolfram.com/language/guide/MachineLearning.en.md) * [Text Manipulation](https://reference.wolfram.com/language/guide/ProcessingTextualData.en.md) * [LLM-Related Functionality](https://reference.wolfram.com/language/guide/LLMFunctions.en.md) * [Text Analysis](https://reference.wolfram.com/language/guide/TextAnalysis.en.md) * [Natural Language Processing](https://reference.wolfram.com/language/guide/NaturalLanguageProcessing.en.md) * [String Manipulation](https://reference.wolfram.com/language/guide/StringManipulation.en.md) * [Working with Templates](https://reference.wolfram.com/language/guide/WorkingWithTemplates.en.md) * [Converting between Expressions & Strings](https://reference.wolfram.com/language/guide/ConvertingBetweenExpressionsAndStrings.en.md) * [Linguistic Data](https://reference.wolfram.com/language/guide/LinguisticData.en.md) * [Free-Form & External Input](https://reference.wolfram.com/language/guide/FreeFormAndExternalInput.en.md) * [Setting Up Input Interpreters](https://reference.wolfram.com/language/guide/InterpretingStrings.en.md) * [Knowledge Representation & Access](https://reference.wolfram.com/language/guide/KnowledgeRepresentationAndAccess.en.md) * [Text Generation](https://reference.wolfram.com/language/guide/TextConstruction.en.md) * [Text Normalization](https://reference.wolfram.com/language/guide/TextNormalization.en.md) ## History * [Introduced in 2023 (13.3)](https://reference.wolfram.com/language/guide/SummaryOfNewFeaturesIn133.en.md) \| [Updated in 2025 (14.3)](https://reference.wolfram.com/language/guide/SummaryOfNewFeaturesIn143.en.md)