API ref for Phi Silica in the Windows App SDK
Important
This feature is not yet available. It is expected to ship in an upcoming experimental channel release of the Windows App SDK.
The Windows App SDK experimental channel includes APIs and features in early stages of development. All APIs in the experimental channel are subject to extensive revisions and breaking changes and may be removed from subsequent releases at any time. They are not supported for use in production environments, and apps that use experimental features cannot be published to the Microsoft Store.
Learn about the Windows App SDK APIs that can access local language models such as Phi Silica, Microsoft's most powerful NPU-tuned local language model that enables on-device processing and generation of chat, reasoning over text, math solving, code generation, and more.
For more details, see Get started with Phi Silica in the Windows App SDK.
Tip
Provide feedback on these APIs and their functionality by creating a new Issue in the Windows App SDK GitHub repo. (Make sure you include Phi Silica in the title!)
Microsoft.Windows.AI.Generative namespace
Provides APIs for local, on-device generative AI prompt processing and responses.
ImageDescriptionGenerator class
public sealed class ImageDescriptionGenerator : System.IDisposable
ImageDescriptionGenerator.Close method
// This member is not implemented in C#
ImageDescriptionGenerator.CreateAsync method
public static Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.ImageDescriptionGenerator> CreateAsync ();
ImageDescriptionGenerator.DescribeAsync(Microsoft.Windows.Imaging.ImageBuffer) method
public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> DescribeAsync (Microsoft.Windows.Imaging.ImageBuffer image);
ImageDescriptionGenerator.IsAvailable method
public static bool IsAvailable ();
ImageDescriptionGenerator.MakeAvailableAsync method
public static Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.Management.Deployment.PackageDeploymentResult,Microsoft.Windows.Management.Deployment.PackageDeploymentProgress> MakeAvailableAsync ();
LanguageModel class
public sealed class LanguageModel : System.IDisposable
Represents an object that can interact with a local language model to generate responses for a provided prompt.
LanguageModel.Close method
Disposes of the object and associated resources.
Remarks
Not implemented in C#.
LanguageModel.CreateAsync method
public static Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModel> CreateAsync ();
Asynchronously creates a new instance of the LanguageModel class.
Returns
A new instance of the TextRecognizer class.
LanguageModel.GenerateEmbeddingVector(System.String) method
public System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> GenerateEmbeddingVector (string prompt);
LanguageModel.GenerateEmbeddingVectorAsync(System.String) method
public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector>> GenerateEmbeddingVectorAsync (string prompt);
LanguageModel.GenerateResponseAsync(System.String) method
public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (string prompt);
Generates and returns a complete response for a single prompt.
Parameters
prompt
A prompt in the form of a question.
Returns
A response string and status.
Exceptions
ArgumentException: The specified prompt is longer than the maximum number of tokens the model can accept.
LanguageModel.GenerateResponseAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,System.String) method
public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt);
LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector}) method
public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromEmbeddingsWithProgressAsync (System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> promptEmbedding);
LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector}) method
public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromEmbeddingsWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> promptEmbedding);
LanguageModel.GenerateResponseWithProgressAsync(System.String) method
public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,
string> GenerateResponseWithProgressAsync (string prompt);
Generates and streams a response through a progress handler. Partial results can be retrieved while generation is in progress.
Parameters
prompt
The prompt for the response.
Returns
A response string and status.
The next token of the string that is being added to the full response as the model returns it, this is the delta from the previous LanguageModelReponse set as the result OnProgress.
Exceptions
ArgumentException: The specified prompt is longer than the maximum number of tokens the model can accept.
Remarks
OnProgress events occur on generation of each single word in the response.
LanguageModel.IsAvailable method
public static bool IsAvailable ();
Retrieves whether the required AI Model is available.
Returns
True, if required AI Model is available. Otherwise, false.
LanguageModel.IsPromptLargerThanContext(System.String) method
public bool IsPromptLargerThanContext (string prompt);
LanguageModel.MakeAvailableAsync method
public static Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.Management.Deployment.PackageDeploymentResult,
Microsoft.Windows.Management.Deployment.PackageDeploymentProgress> MakeAvailableAsync ();
Ensures the underlying language model is installed and available for use.
Returns
An asynchronous action with progress that returns a PackageDeploymentResult on completion.
LanguageModelOptions class
public sealed class LanguageModelOptions
LanguageModelOptions.#ctor constructor
public LanguageModelOptions ();
LanguageModelOptions.#ctor(Microsoft.Windows.AI.Generative.LanguageModelSkill,System.Single,System.Single,System.UInt32) constructor
public LanguageModelOptions (Microsoft.Windows.AI.Generative.LanguageModelSkill skill, float temp, float top_p, uint top_k);
LanguageModelOptions.Skill property
public Microsoft.Windows.AI.Generative.LanguageModelSkill Skill { get; set; }
LanguageModelOptions.Temp property
public float Temp { get; set; }
LanguageModelOptions.Top_k property
public uint Top_k { get; set; }
LanguageModelOptions.Top_p property
public float Top_p { get; set; }
LanguageModelResponse class
public sealed class LanguageModelResponse
Represents a response string and status.
LanguageModelResponse.Response property
public string Response { get; }
Gets the response string returned by the language model based on the provided prompt.
Property value
The response string returned by the language model based on the provided prompt.
LanguageModelResponse.Status property
public Microsoft.Windows.AI.Generative.LanguageModelResponseStatus Status { get; }
Gets the response status based on the provided prompt.
Property value
The response string returned by the language model based on the provided prompt.
Remarks
Any value other than Succeeded or InProgress is considered a failure.
LanguageModelResponseStatus enum
public enum LanguageModelResponseStatus
Specifies the possible response status values for the provided prompt.
Fields - OS
Complete: 0
Response is complete.
InProgress: 1
Response is in progress.
BlockedByPolicy: 2
Response is blocked by a policy setting.
PromptLargerThanContext: 3
LanguageModelSkill enum
public enum LanguageModelSkill