通过

Windows 应用 SDK中 Phi 硅的 API ref

  • 项目

重要

Windows 应用 SDK 的最新试验通道版本中提供。

Windows 应用 SDK 试验通道包括开发早期阶段的 API 和功能。 试验通道中的所有 API 都可能经过大量修订和中断性变更，并且随时可从后续版本中删除。 实验性功能在生产环境中不受支持，并且使用这些功能的应用程序无法发布到 Microsoft Store。

  • 皮硅在中国不可用。
  • 不支持解压缩的应用。

了解可以访问本地语言模型的Windows 应用 SDK API，例如 Phi 硅、Microsoft功能最强大的 NPU 优化本地语言模型，可实现设备处理和聊天生成、推理文本、数学求解、代码生成等。

有关详细信息，请参阅 Windows 应用 SDK 中的 Phi 硅入门。

提示

通过在 Windows 应用 SDK GitHub 存储库中创建 新议题（在标题中包括 Phi Silica），或通过响应 现有议题来提供有关这些 API 及其功能的反馈。

Microsoft.Windows.AI.Generative 命名空间

提供用于本地、设备上的生成式 AI 提示处理和响应的 API。

ImageDescriptionGenerator 类

public sealed class ImageDescriptionGenerator : System.IDisposable

ImageDescriptionGenerator.Close 方法

// This member is not implemented in C#
注解

未在 C# 中实现。

ImageDescriptionGenerator.CreateAsync 方法

public static Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.ImageDescriptionGenerator> CreateAsync ();

ImageDescriptionGenerator.DescribeAsync(Microsoft.Windows.Imaging.ImageBuffer) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> DescribeAsync (Microsoft.Windows.Imaging.ImageBuffer image);

警告

在图像上调用 ImageDescriptionGenerator.DescribeAsync() 时，有时会引发错误。 可以跳过此错误，使调试器能够继续并生成正确的输出。 此错误仅在开发人员环境中可见，不适用于最终用户（使用你的应用的客户）。 使用调试或发布版本将触发此错误。 此错误间歇性地出现，而不是在每次运行时出现。

ImageDescriptionGenerator.DescribeAsync(Microsoft.Graphics.Imaging.ImageBuffer,Microsoft.Windows.AI.Generative.ImageDescriptionScenario) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> DescribeAsync (Microsoft.Graphics.Imaging.ImageBuffer image, Microsoft.Windows.AI.Generative.ImageDescriptionScenario scenario);
参数
image
方案
返回
注解

ImageDescriptionGenerator.DescribeAsync(Microsoft.Graphics.Imaging.ImageBuffer,Microsoft.Windows.AI.Generative.ImageDescriptionScenario,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> DescribeAsync (Microsoft.Graphics.Imaging.ImageBuffer image, Microsoft.Windows.AI.Generative.ImageDescriptionScenario scenario, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
image
方案
内容过滤选项
返回
注解

ImageDescriptionGenerator.IsAvailable 方法

public static bool IsAvailable ();

ImageDescriptionGenerator.MakeAvailableAsync 方法

public static Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.Management.Deployment.PackageDeploymentResult,Microsoft.Windows.Management.Deployment.PackageDeploymentProgress> MakeAvailableAsync ();

ImageDescriptionScenario 枚举器

public enum ImageDescriptionScenario

字段

Accessibility：1
Caption：2
DetailedNarration：3
OfficeCharts：4

ImageLLMAdapterCreator 类

public sealed class ImageLLMAdapterCreator : System.IDisposable

ImageLLMAdapterCreator.Close 方法

// This member is not implemented in C#
注解

未在 C# 中实现。

ImageLLMAdapterCreator.CreateAsync 方法

public static Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.ImageLLMAdapterCreator> CreateAsync ();
返回

ImageLLMAdapterCreator.GetImageLLMEmbeddings(Microsoft.Windows.SemanticSearch.EmbeddingVector) 方法

public System.Collections.Generic.IReadOnlyList<float> GetImageLLMEmbeddings (Microsoft.Windows.SemanticSearch.EmbeddingVector embeddings);
参数
嵌入
返回

ImageLLMAdapterCreator.GetImageLLMEmbeddingsAsync(Microsoft.Windows.SemanticSearch.EmbeddingVector) 方法

public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<float>> GetImageLLMEmbeddingsAsync (Microsoft.Windows.SemanticSearch.EmbeddingVector embeddings);
参数
嵌入
返回

ImageLLMAdapterCreator.GetModelInputSize 方法

public uint GetModelInputSize ();
返回

ImageLLMAdapterCreator.GetModelOutputSize 方法

public uint GetModelOutputSize ();
返回

ImageLLMAdapterCreator.IsAvailable 方法

public static bool IsAvailable ();
返回

ImageLLMAdapterCreator.MakeAvailableAsync 方法

public static Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.Management.Deployment.PackageDeploymentResult,Microsoft.Windows.Management.Deployment.PackageDeploymentProgress> MakeAvailableAsync ();
返回

LanguageModel 类

public sealed class LanguageModel : System.IDisposable

表示可与本地语言模型交互的对象，以便为提供的提示生成响应。

LanguageModel.Close 方法

释放对象和关联的资源。

注解

未在 C# 中实现。

LanguageModel.CreateAsync 方法

public static Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModel> CreateAsync ();

异步创建 LanguageModel 类的新实例。

返回

TextRecognizer 类的新实例。

LanguageModel.CreateContext 方法

public Microsoft.Windows.AI.Generative.LanguageModelContext CreateContext ();
返回
注解

LanguageModel.CreateContext（System.String，Microsoft.Windows.AI.ContentModeration.ContentFilterOptions） 方法

public Microsoft.Windows.AI.Generative.LanguageModelContext CreateContext (string systemPrompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
systemPrompt
内容过滤选项
返回
注解

LanguageModel.GenerateEmbeddingVector(System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> GenerateEmbeddingVector (string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
prompt
内容过滤选项
返回
注解

LanguageModel.GenerateEmbeddingVector(System.String) 方法

public System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> GenerateEmbeddingVector (string prompt);
参数
prompt
返回
注解

LanguageModel.GenerateEmbeddingVectorAsync(System.String) 方法

public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector>> GenerateEmbeddingVectorAsync (string prompt);
参数
prompt
返回
注解

LanguageModel.GenerateEmbeddingVectorAsync(System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector>> GenerateEmbeddingVectorAsync (string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
prompt
内容过滤选项
返回
注解

LanguageModel.GenerateResponseAsync（Microsoft.Windows.AI.Generative.LanguageModelOptions，System.String，Microsoft.Windows.AI.ContentModeration.ContentFilterOptions）方法

public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
选项
prompt
内容过滤选项
返回
注解

LanguageModel.GenerateResponseAsync(System.String) 方法

public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (string prompt);

生成并返回单个提示的完整响应。

参数
prompt

以问题形式显示的提示。

返回

响应字符串和状态。

异常

ArgumentException：指定的提示长于模型可以接受的最大 token 数。

LanguageModel.GenerateResponseAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions,Microsoft.Windows.AI.Generative.LanguageModelContext) 方法

public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions, Microsoft.Windows.AI.Generative.LanguageModelContext context);
参数
选项
prompt
内容过滤选项
上下文
返回
注解

LanguageModel.GenerateResponseAsync（Microsoft.Windows.AI.Generative.LanguageModelOptions，System.String） 方法

public Windows.Foundation.IAsyncOperation<Microsoft.Windows.AI.Generative.LanguageModelResponse> GenerateResponseAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt);
参数
选项
prompt
返回
注解

LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector},Microsoft.Windows.AI.ContentModeration.ContentFilterOptions,Microsoft.Windows.AI.Generative.LanguageModelContext) 方法

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, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions, Microsoft.Windows.AI.Generative.LanguageModelContext context);
参数
选项
promptEmbedding
内容过滤选项
上下文
返回
注解

LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector},Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

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, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
选项
promptEmbedding
内容过滤选项
返回
注解

LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector}) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromEmbeddingsWithProgressAsync (System.Collections.Generic.IReadOnlyList<Microsoft.Windows.SemanticSearch.EmbeddingVector> promptEmbedding);
参数
promptEmbedding
返回
注解

LanguageModel.GenerateResponseFromEmbeddingsWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{Microsoft.Windows.SemanticSearch.EmbeddingVector}) 方法

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);
参数
选项
promptEmbedding
返回

LanguageModel.GenerateResponseFromTokensWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{System.Int64}) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromTokensWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, System.Collections.Generic.IReadOnlyList<long> promptTokens);
参数
选项
promptTokens
返回

LanguageModel.GenerateResponseFromTokensWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{System.Int64},Microsoft.Windows.AI.ContentModeration.ContentFilterOptions,Microsoft.Windows.AI.Generative.LanguageModelContext) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromTokensWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, System.Collections.Generic.IReadOnlyList<long> promptTokens, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions, Microsoft.Windows.AI.Generative.LanguageModelContext context);
参数
选项
promptTokens
内容过滤选项
上下文
返回

LanguageModel.GenerateResponseFromTokensWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,Windows.Foundation.Collections.IVectorView{System.Int64},Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseFromTokensWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, System.Collections.Generic.IReadOnlyList<long> promptTokens, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
选项
promptTokens
内容过滤选项
返回

LanguageModel.GenerateResponseWithProgressAsync（Microsoft.Windows.AI.Generative.LanguageModelOptions，System.String，Microsoft.Windows.AI.ContentModeration.ContentFilterOptions） 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
选项
prompt
内容过滤选项
返回

LanguageModel.GenerateResponseWithProgressAsync(System.String) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseWithProgressAsync (string prompt);
参数
prompt
返回

LanguageModel.GenerateResponseWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,System.String) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt);
参数
选项
prompt
返回

LanguageModel.GenerateResponseWithProgressAsync(Microsoft.Windows.AI.Generative.LanguageModelOptions,System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions,Microsoft.Windows.AI.Generative.LanguageModelContext) 方法

public Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.AI.Generative.LanguageModelResponse,string> GenerateResponseWithProgressAsync (Microsoft.Windows.AI.Generative.LanguageModelOptions options, string prompt, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions, Microsoft.Windows.AI.Generative.LanguageModelContext context);
参数
选项
prompt
内容过滤选项
上下文
返回

LanguageModel.GenerateTokens(System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public System.Collections.Generic.IReadOnlyList<long> GenerateTokens (string text, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
text
内容过滤选项
返回

LanguageModel.GenerateTokens(System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public System.Collections.Generic.IReadOnlyList<long> GenerateTokens (string text, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
text
内容过滤选项
返回

LanguageModel.GenerateTokensAsync(System.String) 方法

public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<long>> GenerateTokensAsync (string text);
参数
text
返回

LanguageModel.GenerateTokensAsync(System.String,Microsoft.Windows.AI.ContentModeration.ContentFilterOptions) 方法

public Windows.Foundation.IAsyncOperation<System.Collections.Generic.IReadOnlyList<long>> GenerateTokensAsync (string text, Microsoft.Windows.AI.ContentModeration.ContentFilterOptions contentFilterOptions);
参数
text
内容过滤选项
返回

LanguageModel.IsAvailable 方法

public static bool IsAvailable ();
返回

LanguageModel.IsPromptLargerThanContext(Microsoft.Windows.AI.Generative.LanguageModelContext,System.String) 方法

public bool IsPromptLargerThanContext (Microsoft.Windows.AI.Generative.LanguageModelContext context, string prompt);
参数
上下文
prompt
返回

LanguageModel.IsPromptLargerThanContext(System.String) 方法

public bool IsPromptLargerThanContext (string prompt);
参数
prompt
返回

LanguageModel.MakeAvailableAsync 方法

public static Windows.Foundation.IAsyncOperationWithProgress<Microsoft.Windows.Management.Deployment.PackageDeploymentResult,Microsoft.Windows.Management.Deployment.PackageDeploymentProgress> MakeAvailableAsync ();
返回

LanguageModelContext 类

public sealed class LanguageModelContext

LanguageModelOptions 类

public sealed class LanguageModelOptions

LanguageModelOptions.#ctor 构造函数

public LanguageModelOptions ();

LanguageModelOptions.#ctor(Microsoft.Windows.AI.Generative.LanguageModelSkill,System.Single,System.Single,System.UInt32) 构造函数

public LanguageModelOptions (Microsoft.Windows.AI.Generative.LanguageModelSkill skill, float temp, float top_p, uint top_k);
参数
技能
temp
top_p
top_k

LanguageModelOptions.Skill 属性

public Microsoft.Windows.AI.Generative.LanguageModelSkill Skill { get; set; }
属性值

LanguageModelOptions.Temp 属性

public float Temp { get; set; }
属性值

LanguageModelOptions.Top_k 属性

public uint Top_k { get; set; }
属性值

LanguageModelOptions.Top_p 属性

public float Top_p { get; set; }
属性值

LanguageModelResponse 类

public sealed class LanguageModelResponse

LanguageModelResponse.#ctor(System.String,Microsoft.Windows.AI.Generative.LanguageModelResponseStatus) 构造函数

public LanguageModelResponse (string response, Microsoft.Windows.AI.Generative.LanguageModelResponseStatus status);
参数
response
status

LanguageModelResponse.Response 属性

public string Response { get; }
属性值

LanguageModelResponse.Status 属性

public Microsoft.Windows.AI.Generative.LanguageModelResponseStatus Status { get; }
属性值

LanguageModelResponseStatus 枚举

public enum LanguageModelResponseStatus

字段

Complete：0
InProgress：1
BlockedByPolicy：2
PromptLargerThanContext： 3
PromptBlockedByPolicy：4
ResponseBlockedByPolicy：5

LanguageModelSkill 枚举

public enum LanguageModelSkill

字段

常规：0
TextToTable：1
汇总：2
重写：3

相关内容