Semantic Kernel:图识文

多模态是每个LLM具有的能力，图片又是最常见的信息载体，GPT对图片的识别也很早就有了，随着GPT版本的迭代，效果越来越好。SK也是在很多就适配了图识文，只不过最近版本才支持本地图片的上传。（有点晚）

图片场景识别：

using Microsoft.SemanticKernel.ChatCompletion;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Connectors.OpenAI;


var chatModelId = "gpt-4o";
var key = File.ReadAllText(@"C:\GPT\key.txt");
#pragma warning disable SKEXP0070
#pragma warning disable SKEXP0010
#pragma warning disable SKEXP0001
#pragma warning disable SKEXP0110
var kernel = Kernel.CreateBuilder()
   .AddOpenAIChatCompletion(chatModelId, key)
   .Build();


var chat = kernel.GetRequiredService<IChatCompletionService>();
var chatHistory = new ChatHistory();
chatHistory.AddUserMessage(new ChatMessageContentItemCollection
{
     new TextContent("请说明这是那里，什么样的天气，大家在干什么?一共有多少人"),
     new ImageContent(File.ReadAllBytes("tam.jpg"),"image/jpeg")
});
var settings = new Dictionary<string, object>
{
    ["max_tokens"] = 1000,
    ["temperature"] = 0.2,
    ["top_p"] = 0.8,
    ["presence_penalty"] = 0.0,
    ["frequency_penalty"] = 0.0
};


var content = chat.GetStreamingChatMessageContentsAsync(chatHistory, new PromptExecutionSettings
{
    ExtensionData = settings
});
await foreach (var item in content)
{
    Console.Write(item.Content);
}
Console.ReadLine();

图片：

结果：

文字识别：

using Microsoft.SemanticKernel.ChatCompletion;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.Connectors.OpenAI;


var chatModelId = "gpt-4o";
var key = File.ReadAllText(@"C:\GPT\key.txt");
#pragma warning disable SKEXP0070
#pragma warning disable SKEXP0010
#pragma warning disable SKEXP0001
#pragma warning disable SKEXP0110
var kernel = Kernel.CreateBuilder()
   .AddOpenAIChatCompletion(chatModelId, key)
   .Build();


var chat = kernel.GetRequiredService<IChatCompletionService>();
var chatHistory = new ChatHistory();
chatHistory.AddUserMessage(new ChatMessageContentItemCollection
{
     new TextContent("请识别图片上的文字，并输出"),
     new ImageContent(File.ReadAllBytes("japancard.png"),"image/jpeg")
});
var settings = new Dictionary<string, object>
{
    ["max_tokens"] = 1000,
    ["temperature"] = 0.2,
    ["top_p"] = 0.8,
    ["presence_penalty"] = 0.0,
    ["frequency_penalty"] = 0.0
};


var content = chat.GetStreamingChatMessageContentsAsync(chatHistory, new PromptExecutionSettings
{
    ExtensionData = settings
});
await foreach (var item in content)
{
    Console.Write(item.Content);
}
Console.ReadLine();

图片：

结果：