English | 简体中文
A non-official DashScope (Bailian) service SDK maintained by Cnblogs.
Note: This project is actively under development. Breaking changes may occur even in minor versions. Please review the Release Notes before upgrading.
Install NuGet package Cnblogs.DashScope.Sdk
var client = new DashScopeClient("your-api-key"); // you could use different api base, example: // var client = new DashScopeClient("your-api-key", "https://dashscope-intl.aliyuncs.com/api/v1/"); var completion = await client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = new List<TextChatMessage>() { TextChatMessage.System("You are a helpful assistant"), TextChatMessage.User("你是谁？") } }, Parameters = new TextGenerationParameters() { ResultFormat = "message" } }); Console.WriteLine(completion.Output.Choices![0].Message.Content)Install NuGet package Cnblogs.DashScope.AspNetCore
Program.cs
builder.AddDashScopeClient(builder.Configuration);appsettings.json
{ "DashScope": { "ApiKey": "your-api-key", "BaseAddress": "https://dashscope-intl.aliyuncs.com/api/v1" } }Application class:
public class YourService(IDashScopeClient client) { public async Task<string> CompletePromptAsync(string prompt) { var completion = await client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = new List<TextChatMessage>() { TextChatMessage.System("You are a helpful assistant"), TextChatMessage.User("你是谁？") } }, Parameters = new TextGenerationParameters() { ResultFormat = "message" } }); return completion.Output.Choices![0].Message.Content } }Install NuGet package Cnblogs.DashScope.AI
var client = new DashScopeClient("your-api-key").AsChatClient("qwen-max"); var completion = await client.GetResponseAsync("hello"); Console.WriteLine(completion.Text);If you need to use input data or parameters not supported by Microsoft.Extensions.AI, you can directly invoke the underlying SDK by passing a raw TextChatMessage or MultimodalMessage via RawPresentation.
Similarly, to pass unsupported parameters, you can also do so directly by setting the raw property within AdditionalProperties.
Example(Using qwen-doc-turbo)
var messages = new List<TextChatMessage>() { TextChatMessage.DocUrl( "从这两份产品手册中，提取所有产品信息，并整理成一个标准的JSON数组。每个对象需要包含：model(产品的型号)、name(产品的名称)、price(价格（去除货币符号和逗号）)", [ "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20251107/jockge/%E7%A4%BA%E4%BE%8B%E4%BA%A7%E5%93%81%E6%89%8B%E5%86%8CA.docx", "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20251107/ztwxzr/%E7%A4%BA%E4%BE%8B%E4%BA%A7%E5%93%81%E6%89%8B%E5%86%8CB.docx" ]) }; var parameters = new TextGenerationParameters() { ResultFormat = "message", IncrementalOutput = true, }; var response = client .AsChatClient("qwen-doc-turbo") .GetStreamingResponseAsync( messages.Select(x => new ChatMessage() { RawRepresentation = x }), new ChatOptions() { AdditionalProperties = new AdditionalPropertiesDictionary() { { "raw", parameters } } }); await foreach (var chunk in response) { Console.Write(chunk.Text); }Similarly, you can also retrieve the raw message from the RawPresentation in the message returned by the model.
Example (Getting the token count for an image when calling qwen3-vl-plus):
var response = client .AsChatClient("qwen3-vl-plus") .GetStreamingResponseAsync( new List<ChatMessage>() { new( ChatRole.User, new List<AIContent>() { new UriContent( "https://help-static-aliyun-doc.aliyuncs.com/file-manage-files/zh-CN/20241022/emyrja/dog_and_girl.jpeg", MediaTypeNames.Image.Jpeg), new UriContent( "https://dashscope.oss-cn-beijing.aliyuncs.com/images/tiger.png", MediaTypeNames.Image.Jpeg), new TextContent("这些图展现了什么内容？") }) }, new ChatOptions()); var lastChunk = (ChatResponseUpdate?)null; await foreach (var chunk in response) { Console.Write(chunk.Text); lastChunk = chunk; } Console.WriteLine(); // Access underlying raw response var raw = lastChunk?.RawRepresentation as ModelResponse<MultimodalOutput, MultimodalTokenUsage>; Console.WriteLine($"Image token usage: {raw?.Usage?.ImageTokens}");- Text Generation - QWen3, DeepSeek, etc. Supports reasoning/tool calling/web search/translation scenarios
- Multimodal - QWen-VL, QVQ, etc. Supports reasoning/visual understanding/OCR/audio understanding
- Text-to-Speech - CosyVoice, Sambert, etc. For TTS applications
- Image Generation - wanx2.1, etc. For text-to-image and portrait style transfer
- Application Call
- Text Embeddings
Use dashScopeClient.GetTextCompletionAsync and dashScopeClient.GetTextCompletionStreamAsync() to access text generation APIs.
Common models: qwen-max qwen-plus qwen-flush etc.
Basic example:
var client = new DashScopeClient("your-api-key"); var completion = await client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = new List<TextChatMessage>() { TextChatMessage.System("You are a helpful assistant"), TextChatMessage.User("Who are you?") } }, Parameters = new TextGenerationParameters() { ResultFormat = "message" } }); Console.WriteLine(completion.Output.Choices![0].Message.Content)The key is maintaining a TextChatMessage array as conversation history.
var messages = new List<TextChatMessage>(); messages.Add(TextChatMessage.System("You are a helpful assistant")); while (true) { Console.Write("User > "); var input = Console.ReadLine(); if (string.IsNullOrEmpty(input)) { Console.WriteLine("Using default input: Who are you?"); input = "Who are you?"; } messages.Add(TextChatMessage.User(input)); var completion = await client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message" } }); Console.WriteLine("Assistant > " + completion.Output.Choices![0].Message.Content); var usage = completion.Usage; if (usage != null) { Console.WriteLine($"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/total({usage.TotalTokens})"); } messages.Add(TextChatMessage.Assistant(completion.Output.Choices[0].Message.Content)); }The model's thinking process is stored in a separate ReasoningContent property. When saving to conversation history, ignore it and only keep the model's reply Content.
Some models accept EnableThinking to control deep thinking, which can be set in Parameters.
var messages = new List<TextChatMessage>(); messages.Add(TextChatMessage.System("You are a helpful assistant")); while (true) { Console.Write("User > "); var input = Console.ReadLine(); if (string.IsNullOrEmpty(input)) { Console.WriteLine("Please enter a user input."); return; } messages.Add(TextChatMessage.User(input)); var completion = await client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true } }); Console.WriteLine("Reasoning > " + completion.Output.Choices![0].Message.ReasoningContent); Console.WriteLine("Assistant > " + completion.Output.Choices![0].Message.Content); var usage = completion.Usage; if (usage != null) { Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/reasoning({usage.OutputTokensDetails?.ReasoningTokens})/total({usage.TotalTokens})"); } messages.Add(TextChatMessage.Assistant(completion.Output.Choices[0].Message.Content)); }var request = new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true, IncrementalOutput = true } }Use client.GetTextCompletionStreamAsync for streaming output. It's recommended to enable IncrementalOutput in Parameters for incremental output.
Incremental output:
Example: ["I love","eating","apples"]
Non-incremental output:
Example: ["I love","I love eating","I love eating apples"]
Streaming output returns an IAsyncEnumerable. Use await foreach to iterate, record incremental content, then save to conversation history.
var messages = new List(); messages.Add(TextChatMessage.System("You are a helpful assistant")); while (true) { Console.Write("User > "); var input = Console.ReadLine(); if (string.IsNullOrEmpty(input)) { Console.WriteLine("Please enter a user input."); return; }
messages.Add(TextChatMessage.User(input)); var completion = client.GetTextCompletionStreamAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true, IncrementalOutput = true } }); var reply = new StringBuilder(); var reasoning = false; TextGenerationTokenUsage? usage = null; await foreach (var chunk in completion) { var choice = chunk.Output.Choices![0]; if (string.IsNullOrEmpty(choice.Message.ReasoningContent) == false) { // reasoning if (reasoning == false) { Console.Write("Reasoning > "); reasoning = true; } Console.Write(choice.Message.ReasoningContent); continue; } if (reasoning) { reasoning = false; Console.WriteLine(); Console.Write("Assistant > "); } Console.Write(choice.Message.Content); reply.Append(choice.Message.Content); usage = chunk.Usage; } Console.WriteLine(); messages.Add(TextChatMessage.Assistant(reply.ToString())); if (usage != null) { Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/reasoning({usage.OutputTokensDetails?.ReasoningTokens})/total({usage.TotalTokens})"); } }
Use ThinkingBudget in Parameters to limit the model's thinking length.
const int budget = 10; Console.WriteLine($"Set thinking budget to {budget} tokens"); var messages = new List<TextChatMessage>(); messages.Add(TextChatMessage.System("You are a helpful assistant")); while (true) { Console.Write("User > "); var input = Console.ReadLine(); if (string.IsNullOrEmpty(input)) { Console.WriteLine("Please enter a user input."); return; } messages.Add(TextChatMessage.User(input)); var completion = client.GetTextCompletionStreamAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true, ThinkingBudget = budget, IncrementalOutput = true } }); var reply = new StringBuilder(); var reasoning = false; TextGenerationTokenUsage? usage = null; await foreach (var chunk in completion) { var choice = chunk.Output.Choices![0]; if (string.IsNullOrEmpty(choice.Message.ReasoningContent) == false) { // reasoning if (reasoning == false) { Console.Write("Reasoning > "); reasoning = true; } Console.Write(choice.Message.ReasoningContent); continue; } if (reasoning) { reasoning = false; Console.WriteLine(); Console.Write("Assistant > "); } Console.Write(choice.Message.Content); reply.Append(choice.Message.Content); usage = chunk.Usage; } Console.WriteLine(); messages.Add(TextChatMessage.Assistant(reply.ToString())); if (usage != null) { Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/reasoning({usage.OutputTokensDetails?.ReasoningTokens})/total({usage.TotalTokens})"); } }Controlled mainly through EnableSearch and SearchOptions in Parameters.
Example request:
var request = new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true, EnableSearch = true, SearchOptions = new TextGenerationSearchOptions() { SearchStrategy = "max", // max/turbo - controls number of search results EnableCitation = true, // Add source citations to model reply CitationFormat = "[ref_<number>]", // Citation format EnableSource = true, // Return search source list in SearchInfo ForcedSearch = true, // Force model to search EnableSearchExtension = true, // Enable vertical domain search, results in SearchInfo.Extra PrependSearchResult = true // First packet contains only search results } } };Provide available tools to the model via Tools in Parameters. The model returns messages with ToolCall properties to invoke tools.
After receiving the message, the server needs to call the corresponding tools and insert results as Tool role messages into the conversation history before making another request. The model will summarize the tool call results.
By default, the model calls tools once per turn. To enable multiple tool calls, enable ParallelToolCalls in Parameters.
var tools = new List<ToolDefinition> { new( ToolTypes.Function, new FunctionDefinition( nameof(GetWeather), "Get current weather", new JsonSchemaBuilder().FromType<WeatherReportParameters>().Build())) }; var request = new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = true, IncrementalOutput = true, Tools = tools, ToolChoice = ToolChoice.AutoChoice, ParallelToolCalls = true // Model can call multiple tools at once } } Set ResponseFormat (remarks: not ResultFormat) in Parameters to JSON to force JSON output.
var request = new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-plus", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", ResponseFormat = DashScopeResponseFormat.Json, IncrementalOutput = true } }Place the prefix to complete as an assistant message at the end of the messages array with Partial set to true. The model will complete the remaining content based on this prefix.
Deep thinking cannot be enabled in this mode.
var messages = new List<TextChatMessage> { TextChatMessage.User("Complete following C# method."), TextChatMessage.Assistant("public int Fibonacci(int n)", partial: true) }; var completion = client.GetTextCompletionStreamAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-turbo", Input = new TextGenerationInput() { Messages = messages }, Parameters = new TextGenerationParameters() { ResultFormat = "message", IncrementalOutput = true } });For Qwen-Long models:
var file = new FileInfo("test.txt"); var uploadedFile = await dashScopeClient.OpenAiCompatibleUploadFileAsync(file.OpenRead(), file.Name); var history = new List<ChatMessage> { ChatMessage.File(uploadedFile.Id) }; var completion = await client.client.GetTextCompletionAsync( new ModelRequest<TextGenerationInput, ITextGenerationParameters>() { Model = "qwen-long", Input = new TextGenerationInput() { Messages = history }, Parameters = new TextGenerationParameters() { ResultFormat = "message", EnableThinking = false, } }); Console.WriteLine(completion.Output.Choices[0].Message.Content); // Cleanup await dashScopeClient.OpenAiCompatibleDeleteFileAsync(uploadedFile.Id);Use GetMultimodalGenerationAsync/GetMultimodalGenerationStreamAsync Official Documentation
You can upload file to get an oss link before multimodal usage.
await using var video = File.OpenRead("sample.mp4"); var ossLink = await client.UploadTemporaryFileAsync("qwen3-vl-plus", video, "sample.mp4"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage>(); messages.Add( MultimodalMessage.User( [ MultimodalMessageContent.VideoContent(ossLink, fps: 2), // MultimodalMessageContent.VideoFrames(links), // MultimodalMessageContent.ImageContent(link) MultimodalMessageContent.TextContent("这段视频的内容是什么？") ]));var image = await File.ReadAllBytesAsync("Lenna.jpg"); var response = dashScopeClient.GetMultimodalGenerationStreamAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qvq-plus", Input = new MultimodalInput() { Messages = [ MultimodalMessage.User( [ MultimodalMessageContent.ImageContent(image, "image/jpeg"), MultimodalMessageContent.TextContent("她是谁？") ]) ] }, Parameters = new MultimodalParameters { IncrementalOutput = true, // EnableThinking = true, VlHighResolutionImages = false } }); // output var reasoning = false; await foreach (var modelResponse in response) { var choice = modelResponse.Output.Choices.FirstOrDefault(); if (choice != null) { if (choice.FinishReason != "null") { break; } if (string.IsNullOrEmpty(choice.Message.ReasoningContent) == false) { if (reasoning == false) { reasoning = true; Console.WriteLine("<think>"); } Console.Write(choice.Message.ReasoningContent); continue; } if (reasoning) { reasoning = false; Console.WriteLine("</think>"); } Console.Write(choice.Message.Content[0].Text); } }Base example of OCR
// upload file await using var tilted = File.OpenRead("tilted.png"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", tilted, "tilted.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage>(); messages.Add( MultimodalMessage.User( [ // set enableRotate to true if your source image is tilted. MultimodalMessageContent.ImageContent(ossLink, enableRotate: true), ])); var completion = client.GetMultimodalGenerationStreamAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput() { Messages = messages }, Parameters = new MultimodalParameters() { IncrementalOutput = true, } }); var reply = new StringBuilder(); var first = false; MultimodalTokenUsage? usage = null; await foreach (var chunk in completion) { var choice = chunk.Output.Choices[0]; if (first) { first = false; Console.Write("Assistant > "); } if (choice.Message.Content.Count == 0) { continue; } Console.Write(choice.Message.Content[0].Text); reply.Append(choice.Message.Content[0].Text); usage = chunk.Usage; } Console.WriteLine(); messages.Add(MultimodalMessage.Assistant([MultimodalMessageContent.TextContent(reply.ToString())])); if (usage != null) { Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/image({usage.ImageTokens})/total({usage.TotalTokens})"); }When using this task, do not enable streaming. Otherwise, completion.Output.Choices[0].Message.Content[0].OcrResult.WordsInfo will be null.
In addition to the standard text content, this task also returns the coordinates of the text.
To call this built-in task, set Parameters.OcrOptions.Task to advanced_recognition. No additional prompt is required.
var messages = new List<MultimodalMessage>(); messages.Add( MultimodalMessage.User( [ MultimodalMessageContent.ImageContent(ossLink), ])); var completion = client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput() { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "advanced_recognition" } } });Output usage:
Console.WriteLine("Text:"); Console.WriteLine(completion.Output.Choices[0].Message.Content[0].Text); Console.WriteLine("WordsInfo:"); foreach (var info in completion.Output.Choices[0].Message.Content[0].OcrResult!.WordsInfo!) { var location = $"[{string.Join(',', info.Location)}]"; var rect = $"[{string.Join(',', info.RotateRect)}]"; Console.WriteLine(info.Text); Console.WriteLine($"Location: {location}"); Console.WriteLine($"RotateRect: {rect}"); Console.WriteLine(); }Output:
Text: ```json [ {"rotate_rect": [236, 254, 115, 299, 90], "text": "OpenAI 兼容"}, {"rotate_rect": [646, 254, 115, 269, 90], "text": "DashScope"}, {"rotate_rect": [236, 684, 115, 163, 90], "text": "Python"}, {"rotate_rect": [492, 684, 115, 105, 90], "text": "Java"}, {"rotate_rect": [712, 684, 115, 85, 90], "text": "curl"} ] ``` WordsInfo: OpenAI 兼容 Location: [46,55,205,55,205,87,46,87] RotateRect: [125,71,159,32,0] DashScope Location: [272,55,415,55,415,87,272,87] RotateRect: [344,71,32,143,90] Python Location: [82,175,169,175,169,207,82,207] RotateRect: [126,191,32,87,90] Java Location: [234,175,289,175,289,207,234,207] RotateRect: [262,191,55,32,0] curl Location: [356,175,401,175,401,207,356,207] RotateRect: [378,191,32,45,90]When using built-in tasks, it is not recommended to enable streaming; otherwise, completion.Output.Choices[0].Message.Content[0].OcrResult.KvResult will be null.
To invoke this built-in task, set Parameters.OcrOptions.Task to key_information_extraction. No additional text information needs to be provided.
You can customize the output JSON format via Parameters.OcrOptions.TaskConfig.ResultSchema (with a maximum of 3 levels of nesting). If left blank, all fields will be output by default.
For example, suppose we want to extract objects of the following type from an image (JSON property names should, as much as possible, be based on the text present in the image):
internal class ReceiptModel() { [JsonPropertyName("乘车日期")] public string? Date { get; init; } [JsonPropertyName("发票")] public ReceiptSerials? Serials { get; init; } } internal class ReceiptSerials { [JsonPropertyName("发票代码")] public string? Code { get; init; } [JsonPropertyName("发票号码")] public string? Serial { get; init; } }Property could be null if model failed to extract value for it. Example request:
await using var file = File.OpenRead("receipt.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "receipt.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput() { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "key_information_extraction", TaskConfig = new MultimodalOcrTaskConfig() { ResultSchema = new Dictionary<string, object>() { { "发票", new Dictionary<string, string>() { { "发票代码", "提取图中的发票代码，通常为一组数字或字母组合" }, { "发票号码", "提取发票上的号码，通常由纯数字组成。" } } }, { "乘车日期", "对应图中乘车日期时间，格式为年-月-日，比如2025-03-05" } } } } } });Consume:
KvResult is JsonElement，you can deserialize it to any type you desire, or you could just use Dictionary<string, JsonElement?> .
Console.WriteLine("Text:"); Console.WriteLine(completion.Output.Choices[0].Message.Content[0].Text); Console.WriteLine("KvResults:"); var model = completion.Output.Choices[0].Message.Content[0].OcrResult!.KvResult?.Deserialize<ReceiptModel>(); Console.WriteLine($"Date: {model?.Date}"); Console.WriteLine($"Code: {model?.Serials?.Code}"); Console.WriteLine($"Serial: {model?.Serials?.Serial}"); /* Text: ```json { "乘车日期": "2013-06-29", "发票": { "发票代码": "221021325353", "发票号码": "10283819" } } ``` KvResults: Date: 2013-06-29 Code: 221021325353 Serial: 10283819 Usage: in(524)/out(65)/image(310)/total(589) */To invoke this built-in task, set Parameters.OcrOptions.Task to table_parsing. No additional text information needs to be provided.
This task will extract tables from images and return them in HTML format.
Example:
await using var file = File.OpenRead("table.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "table.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "table_parsing", } } }); Console.WriteLine(completion.Output.Choices[0].Message.Content[0].Text);Return value(Model would wrap the html code in a markdown code fence):
```html <table> <tr> <td>Record of test data</td> </tr> <tr> <td>Project name：2B</td> <td>Control No.CEPRI-D-JS1-JS-057-2022-003</td> </tr> <tr> <td>Case name</td> <td>Test No.3 Conductor rupture GL+GR(max angle)</td> <td>Last load grade：</td> <td>0%</td> <td>Current load grade：</td> </tr> <tr> <td>Measure</td> <td>Load point</td> <td>Load method</td> <td>Actual Load(%)</td> <td>Actual Load(kN)</td> </tr> <tr> <td>channel</td> <td>V1</td> <td>活载荷</td> <td>147.95</td> <td>0.815</td> </tr> <tr> <td>V03</td> <td>V2</td> <td>活载荷</td> <td>111.75</td> <td>0.615</td> </tr> <tr> <td>V04</td> <td>V3</td> <td>活载荷</td> <td>9.74</td> <td>1.007</td> </tr> <tr> <td>V05</td> <td>V4</td> <td>活载荷</td> <td>7.88</td> <td>0.814</td> </tr> <tr> <td>V06</td> <td>V5</td> <td>活载荷</td> <td>8.11</td> <td>0.780</td> </tr> <tr> <td>V07</td> <td>V6</td> <td>活载荷</td> <td>8.54</td> <td>0.815</td> </tr> <tr> <td>V08</td> <td>V7</td> <td>活载荷</td> <td>6.77</td> <td>0.700</td> </tr> <tr> <td>V09</td> <td>V8</td> <td>活载荷</td> <td>8.59</td> <td>0.888</td> </tr> <tr> <td>L01</td> <td>L1</td> <td>活载荷</td> <td>13.33</td> <td>3.089</td> </tr> <tr> <td>L02</td> <td>L2</td> <td>活载荷</td> <td>9.69</td> <td>2.247</td> </tr> <tr> <td>L03</td> <td>L3</td> <td></td> <td>2.96</td> <td>1.480</td> </tr> <tr> <td>L04</td> <td>L4</td> <td></td> <td>3.40</td> <td>1.700</td> </tr> <tr> <td>L05</td> <td>L5</td> <td></td> <td>2.45</td> <td>1.224</td> </tr> <tr> <td>L06</td> <td>L6</td> <td></td> <td>2.01</td> <td>1.006</td> </tr> <tr> <td>L07</td> <td>L7</td> <td></td> <td>2.38</td> <td>1.192</td> </tr> <tr> <td>L08</td> <td>L8</td> <td></td> <td>2.10</td> <td>1.050</td> </tr> <tr> <td>T01</td> <td>T1</td> <td>活载荷</td> <td>25.29</td> <td>3.073</td> </tr> <tr> <td>T02</td> <td>T2</td> <td>活载荷</td> <td>27.39</td> <td>3.327</td> </tr> <tr> <td>T03</td> <td>T3</td> <td>活载荷</td> <td>8.03</td> <td>2.543</td> </tr> <tr> <td>T04</td> <td>T4</td> <td>活载荷</td> <td>11.19</td> <td>3.542</td> </tr> <tr> <td>T05</td> <td>T5</td> <td>活载荷</td> <td>11.34</td> <td>3.592</td> </tr> <tr> <td>T06</td> <td>T6</td> <td>活载荷</td> <td>16.47</td> <td>5.217</td> </tr> <tr> <td>T07</td> <td>T7</td> <td>活载荷</td> <td>11.05</td> <td>3.498</td> </tr> <tr> <td>T08</td> <td>T8</td> <td>活载荷</td> <td>8.66</td> <td>2.743</td> </tr> <tr> <td>T09</td> <td>WT1</td> <td>活载荷</td> <td>36.56</td> <td>2.365</td> </tr> <tr> <td>T10</td> <td>WT2</td> <td>活载荷</td> <td>24.55</td> <td>2.853</td> </tr> <tr> <td>T11</td> <td>WT3</td> <td>活载荷</td> <td>38.06</td> <td>4.784</td> </tr> <tr> <td>T12</td> <td>WT4</td> <td>活载荷</td> <td>37.70</td> <td>5.030</td> </tr> <tr> <td>T13</td> <td>WT5</td> <td>活载荷</td> <td>30.48</td> <td>4.524</td> </tr> </table> ```To invoke this built-in task, set Parameters.OcrOptions.Task to document_parsing. No additional text information needs to be provided.
This task read images(usually scanned PDF) and return them in LaTeX format.
Example:
await using var file = File.OpenRead("scanned.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "scanned.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "document_parsing", } } });Returns:
```latex \section*{Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution} Peng Wang* \quad Shuai Bai* \quad Sinan Tan* \quad Shijie Wang* \quad Zhihao Fan* \quad Jinze Bai*? \\ Keqin Chen \quad Xuejing Liu \quad Jialin Wang \quad Wenbin Ge \quad Yang Fan \quad Kai Dang \quad Mengfei Du \\ Xuancheng Ren \quad Rui Men \quad Dayiheng Liu \quad Chang Zhou \quad Jingren Zhou \quad Junyang Lin*? \\ Qwen Team \quad Alibaba Group \begin{abstract} We present the Qwen2-VL Series, an advanced upgrade of the previous Qwen-VL models that redefines the conventional predetermined-resolution approach in visual processing. Qwen2-VL introduces the Naive Dynamic Resolution mechanism, which enables the model to dynamically process images of varying resolutions into different numbers of visual tokens. This approach allows the model to generate more efficient and accurate visual representations, closely aligning with human perceptual processes. The model also integrates Multimodal Rotary Position Embedding (M-RoPE), facilitating the effective fusion of positional information across text, images, and videos. We employ a unified paradigm for processing both images and videos, enhancing the model's visual perception capabilities. To explore the potential of large multimodal models, Qwen2-VL investigates the scaling laws for large vision-language models (LVLMS). By scaling both the model size-with versions at 2B, 8B, and 72B parameters-and the amount of training data, the Qwen2-VL Series achieves highly competitive performance. Notably, the Qwen2-VL-72B model achieves results comparable to leading models such as GPT-4o and Claude3.5-Sonnet across various multimodal benchmarks, outperforming other generalist models. Code is available at \url{https://github.com/QwenLM/Qwen2-VL}. \end{abstract} \section{Introduction} In the realm of artificial intelligence, Large Vision-Language Models (LVLMS) represent a significant leap forward, building upon the strong textual processing capabilities of traditional large language models. These advanced models now encompass the ability to interpret and analyze a broader spectrum of data, including images, audio, and video. This expansion of capabilities has transformed LVLMS into indispensable tools for tackling a variety of real-world challenges. Recognized for their unique capacity to condense extensive and intricate knowledge into functional representations, LVLMS are paving the way for more comprehensive cognitive systems. By integrating diverse data forms, LVLMS aim to more closely mimic the nuanced ways in which humans perceive and interact with their environment. This allows these models to provide a more accurate representation of how we engage with and perceive our environment. Recent advancements in large vision-language models (LVLMS) (Li et al., 2023c; Liu et al., 2023b; Dai et al., 2023; Zhu et al., 2023; Huang et al., 2023a; Bai et al., 2023b; Liu et al., 2023a; Wang et al., 2023b; OpenAI, 2023; Team et al., 2023) have led to significant improvements in a short span. These models (OpenAI, 2023; Touvron et al., 2023a,b; Chiang et al., 2023; Bai et al., 2023a) generally follow a common approach of \textit{visual encoder} $\rightarrow$ \textit{cross-modal connector} $\rightarrow$ \textit{LLM}. This setup, combined with next-token prediction as the primary training method and the availability of high-quality datasets (Liu et al., 2023a; Zhang et al., 2023; Chen et al., 2023b); *Equal core contribution, ?Corresponding author ```To invoke this built-in task, set Parameters.OcrOptions.Task to formula_recognition. No additional text information needs to be provided.
This task read images(like handwriting formulas) and return them in LaTeX format.
Example:
// upload file await using var file = File.OpenRead("math.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "math.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "formula_recognition", } } }); Console.WriteLine("LaTeX:"); Console.WriteLine(completion.Output.Choices[0].Message.Content[0].Text); if (completion.Usage != null) { var usage = completion.Usage; Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/image({usage.ImageTokens})/total({usage.TotalTokens})"); }Returns:
```latex \begin{align*} \tilde{G}(x) &= \frac{\alpha}{\kappa}x, \quad \tilde{T}_i = T, \quad \tilde{H}_i = \tilde{\kappa}T, \quad \tilde{\lambda}_i = \frac{1}{\kappa}\sum_{j=1}^{m}\omega_j - z_i, \\ L(\{p_n\}; m^n) + L(\{x^n\}, m^n) + L(\{m^n\}; q_n) &= L(m^n; q_n) \\ I^{m_n} - (L+1) &= z + \int_0^1 I^{m_n} - (L)z \leq x_m | L^{m_n} - (L) |^3 \\ &\leq \kappa\partial_1\psi(x) + \frac{\kappa^3}{6}\partial_2^3\psi(x) - V(x) \psi(x) = \int d^3y K(x,y) \psi(y), \\ \int_{B_{\kappa}(0)} I^{m}(w)^2 d\gamma &= \lim_{n\to\infty} \int_{B_{\kappa}(0)} r\psi(w_n)^2 d\gamma = \lim_{n\to\infty} \int_{B_{\kappa}(y_n)} d\gamma \geq \beta > 0, \end{align*} ```To invoke this built-in task, set Parameters.OcrOptions.Task to text_recognition. No additional text information needs to be provided.
This task read the images and returns content in plain text(Chinese and English only).
Example:
// upload file await using var file = File.OpenRead("webpage.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "webpage.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "text_recognition", } } });Returns
OpenAI 兼容 DashScope Python Java curl To invoke this built-in task, set Parameters.OcrOptions.Task to multi_lan. No additional text information needs to be provided.
This task read the images and returns content in plain text(Support more languages).
await using var file = File.OpenRead("multilanguage.jpg"); var ossLink = await client.UploadTemporaryFileAsync("qwen-vl-ocr-latest", file, "multilanguage.jpg"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User([MultimodalMessageContent.ImageContent(ossLink)]) }; var completion = await client.GetMultimodalGenerationAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen-vl-ocr-latest", Input = new MultimodalInput { Messages = messages }, Parameters = new MultimodalParameters() { OcrOptions = new MultimodalOcrOptions() { Task = "multi_lan", } } });Returns
INTERNATIONAL MOTHER LANGUAGE DAY 你好! Прив?т! Bonjour! Merhaba! Ciao! Hello! Ola! ???? Salam!Use gui-plus to generate standardized operational information based on screenshots and user intent.
Currently, the main capabilities are implemented through the System Prompt, where you can configure the model’s available capabilities and the output JSON format.
Sample system prompt:
## 1. Core Role You are an expert AI Vision Operation Agent. Your task is to analyze computer screenshots, understand user instructions, and then break down the task into single, precise GUI atomic operations. ## 2. [CRITICAL] JSON Schema & Absolute Rules Your output **must** be a JSON object that strictly adheres to the following rules. **Any deviation will result in failure**. - **[R1] Strict JSON**: Your response **must** be *and only be* a JSON object. Do not add any text, comments, or explanations before or after the JSON code block. - **[R2] Strict `thought` Structure**: The `thought` object must contain a single sentence briefly describing your thought process. For example: "The user wants to open the browser. I see the Chrome icon on the desktop, so the next step is to click it." - **[R3] Precise `action` Value**: The value of the `action` field **must** be an uppercase string defined in `## 3. Toolset` (e.g., `"CLICK"`, `"TYPE"`). No leading/trailing spaces or case variations are allowed. - **[R4] Strict `parameters` Structure**: The structure of the `parameters` object **must** be **perfectly identical** to the template defined for the selected Action in `## 3. Toolset`. Key names and value types must match exactly. ## 3. Toolset (Available Actions) ### CLICK - **Description**: Click on the screen. - **Parameters Template**: { "x": <integer>, "y": <integer>, "description": "<string, optional: A short string describing what you are clicking on, e.g., 'Chrome browser icon' or 'Login button'.>" } ### TYPE - **Description**: Type text. - **Parameters Template**: { "text": "<string>", "needs_enter": <boolean> } ### SCROLL - **Description**: Scroll the window. - **Parameters Template**: { "direction": "<'up' or 'down'>", "amount": "<'small', 'medium', or 'large'>" } ### KEY_PRESS - **Description**: Press a function key. - **Parameters Template**: { "key": "<string: e.g., 'enter', 'esc', 'alt+f4'>" } ### FINISH - **Description**: Task completed successfully. - **Parameters Template**: { "message": "<string: A summary of the task completion>" } ### FAILE - **Description**: Task cannot be completed. - **Parameters Template**: { "reason": "<string: A clear explanation of the failure reason>" } ## 4. Thinking and Decision Framework Before generating each action, strictly follow the following thought-verification process: **Goal Analysis**: What is the user's ultimate goal? **Screen Observation (Grounded Observation)**: Carefully analyze the screenshot. Your decisions must be based on visual evidence present in the screenshot. If you cannot see an element, you cannot interact with it. **Action Decision**: Based on the goal and visible elements, select the most appropriate tool. **Construct Output**: a. Record your thought process in the `thought` field. b. Select an `action`. c. Precisely copy the `parameters` template for that action and fill in the values. **Final Verification (Self-Correction)**: Before outputting, perform a final check: - Is my response pure JSON? - Is the `action` value correct (uppercase, no spaces)? - Is the `parameters` structure 100% identical to the template? For example, for `CLICK`, are there separate `x` and `y` keys, and are their values integers?Request:
var messages = new List<MultimodalMessage> { MultimodalMessage.System([MultimodalMessageContent.TextContent(SystemPrompt)]), MultimodalMessage.User( [ MultimodalMessageContent.ImageContent( "https://img.alicdn.com/imgextra/i2/O1CN016iJ8ob1C3xP1s2M6z_!!6000000000026-2-tps-3008-1758.png"), MultimodalMessageContent.TextContent("Open browser") ]) }; var completion = client.GetMultimodalGenerationStreamAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "gui-plus", Input = new MultimodalInput() { Messages = messages }, Parameters = new MultimodalParameters() { IncrementalOutput = true, } });Response:
{ "thought": "用户想打开浏览器，我看到了桌面上的Google Chrome图标，因此下一步是点击它。", "action": "CLICK", "parameters": { "x": 1089, "y": 123 } }Then you can execute the command that model returns, and reply the screenshot with next intension.
Example(use Qwen3-Omni-Captioner)
// upload file await using var audio = File.OpenRead("noise.wav"); var ossLink = await client.UploadTemporaryFileAsync("qwen3-omni-30b-a3b-captioner", audio, "noise.wav"); Console.WriteLine($"File uploaded: {ossLink}"); var messages = new List<MultimodalMessage> { MultimodalMessage.User( [ // 也可以直接传入公网地址 MultimodalMessageContent.AudioContent(ossLink), ]) }; var completion = client.GetMultimodalGenerationStreamAsync( new ModelRequest<MultimodalInput, IMultimodalParameters>() { Model = "qwen3-omni-30b-a3b-captioner", Input = new MultimodalInput() { Messages = messages }, Parameters = new MultimodalParameters() { IncrementalOutput = true, } }); var reply = new StringBuilder(); var first = true; MultimodalTokenUsage? usage = null; await foreach (var chunk in completion) { var choice = chunk.Output.Choices[0]; if (first) { first = false; Console.WriteLine(); Console.Write("Assistant > "); } if (choice.Message.Content.Count == 0) { continue; } Console.Write(choice.Message.Content[0].Text); reply.Append(choice.Message.Content[0].Text); usage = chunk.Usage; } Console.WriteLine(); messages.Add(MultimodalMessage.Assistant([MultimodalMessageContent.TextContent(reply.ToString())])); if (usage != null) { Console.WriteLine( $"Usage: in({usage.InputTokens})/out({usage.OutputTokens})/audio({usage.InputTokensDetails?.AudioTokens})/total({usage.TotalTokens})"); }Sample output
Assistant > The audio clip opens with a rapid, percussive metallic clatter, reminiscent of a typewriter or similar mechanical device, which continues in a steady rhythm throughout the recording. This clatter is slightly left-of-center in the stereo field and is accompanied by a faint, low-frequency hum, likely from a household appliance or HVAC system. The acoustic environment is a small, enclosed room with hard surfaces, indicated by the short, bright reverberation of both the clatter and the speaker’s voice. The audio quality is moderate, with a noticeable electronic hiss and some loss of high-frequency detail, but no digital distortion or clipping. At the one-second mark, a male voice enters, positioned slightly right-of-center and closer to the microphone. He speaks in standard Mandarin, with a tone of weary exasperation: “哎 呀，这样我还怎么安静工作啊？” (“Aiyā, zěnyàng wǒ hái zěnme ānjìng gōngzuò a?”), which translates to “Oh, how can I possibly work quietly like this?” His speech is clear, with a slightly rising pitch on “安静” (“quietly”) and a falling pitch on “啊” (“a”), conveying a sense of complaint and fatigue. The accent is standard, with no regional inflection, and the voice is that of a young to middle-aged adult male. Throughout the clip, the mechanical clatter remains constant and prominent, occasionally competing with the voice for clarity. There are no other sounds, such as footsteps, additional voices, or environmental noises, and the background is otherwise quiet. The interplay between the persistent mechanical noise and the speaker’s complaint creates a vivid sense of disruption and frustration, suggesting an environment where work is being impeded by an external, uncontrolled sound source. Culturally, the use of Mandarin, standard pronunciation, and modern recording quality indicate a contemporary, urban Chinese setting. The language and tone are universally relatable, reflecting a common experience of being disturbed during work. The lack of regional markers or distinctive background noises suggests a generic, possibly domestic or office-like space, but with no clear indicators of a specific location or social context. In summary, the audio portrays a modern Mandarin-speaking man, exasperated by a constant, distracting mechanical noise (likely a typewriter or similar device), attempting to work in a small, reverberant room. The recording’s technical and acoustic features reinforce the sense of disruption and frustration, while the language and setting suggest a contemporary, urban Chinese context. Usage: in(160)/out(514)/audio(152)/total(674)Create a speech synthesis session using dashScopeClient.CreateSpeechSynthesizerSocketSessionAsync().
Note: Use the using statement to automatically dispose the session, or manually call Dispose() to release resources. Avoid reusing sessions.
Create a synthesis session:
using var tts = await dashScopeClient.CreateSpeechSynthesizerSocketSessionAsync("cosyvoice-v2"); var taskId = await tts.RunTaskAsync(new SpeechSynthesizerParameters { Voice = "longxiaochun_v2", Format = "mp3" }); await tts.ContinueTaskAsync(taskId, "Cnblogs"); await tts.ContinueTaskAsync(taskId, "Code changes the world"); await tts.FinishTaskAsync(taskId); var file = new FileInfo("tts.mp3"); using var stream = file.OpenWrite(); await foreach (var b in tts.GetAudioAsync()) { stream.WriteByte(b); } Console.WriteLine($"Audio saved to {file.FullName}");Use shortcuts for Wanx models:
var task = await dashScopeClient.CreateWanxImageSynthesisTaskAsync( WanxModel.WanxV21Turbo, "A futuristic cityscape at sunset", new ImageSynthesisParameters { Style = ImageStyles.OilPainting }); // Pull status while (true) { var result = await dashScopeClient.GetWanxImageSynthesisTaskAsync(task.TaskId); if (result.Output.TaskStatus == DashScopeTaskStatus.Succeeded) { Console.WriteLine($"Image URL: {result.Output.Results[0].Url}"); break; } await Task.Delay(500); }Use CreateWanxImageGenerationTaskAsync and GetWanxImageGenerationTaskAsync
Use CreateWanxBackgroundGenerationTaskAsync and GetWanxBackgroundGenerationTaskAsync
var request = new ApplicationRequest() { Input = new ApplicationInput() { Prompt = "Summarize this file." }, Parameters = new ApplicationParameters() { TopK = 100, TopP = 0.8f, Seed = 1234, Temperature = 0.85f, RagOptions = new ApplicationRagOptions() { PipelineIds = ["thie5bysoj"], FileIds = ["file_d129d632800c45aa9e7421b30561f447_10207234"] } } }; var response = await client.GetApplicationResponseAsync("your-application-id", request); Console.WriteLine(response.Output.Text);ApplicationRequest uses Dictionary<string, object?> as the default type for BizParams.
var request = new ApplicationRequest() { Input = new ApplicationInput() { Prompt = "Summarize this file.", BizParams = new Dictionary<string, object?>() { { "customKey1", "custom-value" } } } }; var response = await client.GetApplicationResponseAsync("your-application-id", request); Console.WriteLine(response.Output.Text);For strong typing support, you can use the generic class ApplicationRequest<TBizParams>. Note that the SDK uses snake_case for JSON serialization. If your application uses different naming conventions, manually specify the serialized property names using [JsonPropertyName("camelCase")].
public record TestApplicationBizParam( [property: JsonPropertyName("sourceCode")] string SourceCode); var request = new ApplicationRequest<TestApplicationBizParam>() { Input = new ApplicationInput<TestApplicationBizParam>() { Prompt = "Summarize this file.", BizParams = new TestApplicationBizParam("test") } }; var response = await client.GetApplicationResponseAsync("your-application-id", request); Console.WriteLine(response.Output.Text);var text = "Sample text for embedding"; var response = await dashScopeClient.GetTextEmbeddingsAsync( TextEmbeddingModel.TextEmbeddingV4, [text], new TextEmbeddingParameters { Dimension = 512 }); var embedding = response.Output.Embeddings.First().Embedding; Console.WriteLine($"Embedding vector length: {embedding.Length}");See Snapshot Files for API parameter examples.
Review Tests for comprehensive usage examples.