Skip to content

Commit

Permalink
Updating readme for SK API
Browse files Browse the repository at this point in the history
  • Loading branch information
@MarkWme
MarkWme committed Oct 12, 2023
1 parent 4e7ea51 commit 4d6e461
Showing 1 changed file with 96 additions and 63 deletions.
159 changes: 96 additions & 63 deletions labs/04-deploy-ai/01-acs-sk-csharp-api/README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6,7 +6,7 @@ The entire solution is in this folder, but we also have all the step by step ins

## Complete Solution

1. To test locally fill in appsettings.json with the same values from the .env file that was used for the Jupyter Notebooks.
1. To test locally fill in `appsettings.json` with the same values from the .env file that was used for the Jupyter Notebooks.
2. Build and Run the App

```csharp
Expand Down Expand Up @@ -38,21 +38,20 @@ dotnet add acs-sk-csharp.csproj package Azure.Search.Documents --version "11.5.0

### Update Program.cs

The first thing we need to do is to add some Using statements.
The first thing we need to do is to add some Using statements to the `Program.cs` file.

```csharp
using System.Text;
using Microsoft.SemanticKernel;
using Microsoft.SemanticKernel.SemanticFunctions;
using Microsoft.SemanticKernel.Orchestration;
using Microsoft.SemanticKernel.SkillDefinition;
using Azure;
using Azure.Search.Documents;
using Azure.Search.Documents.Models;
using Azure.AI.OpenAI;
using Microsoft.AspNetCore.Mvc;
```

Next we read in the variables. Because this is ASP.NET Core we will switch from the DotEnv package to native ASP.NET Core Configuration.
Next we read in the variables. Because this is ASP.NET Core we will switch from the DotEnv package we used in earlier labs to native ASP.NET Core Configuration.

```csharp
var builder = WebApplication.CreateBuilder(args);
Expand Down Expand Up @@ -94,7 +93,7 @@ builder.Services.AddScoped(sp =>
});
```

Next we swap out the app.MapGet weatherforecast function with our own.
Next we swap out the `app.MapGet` `weatherforecast` function with our own.

```csharp
// Configure Routing
Expand All @@ -114,59 +113,10 @@ app.MapPost("/completion", async ([FromServices] IKernel kernel, [FromBody] Comp
// The question is being passed in via the message body.
// [FromBody] string question
// Create a prompt template with variables, note the double curly braces with dollar sign for the variables
// First let's create the prompt string.
var sk_prompt = @"
Question: {{$original_question}}
Do not use any other data.
Only use the movie data below when responding.
{{$search_results}}
";
// Create the PromptTemplateConfig
PromptTemplateConfig promptConfig = new PromptTemplateConfig
{
Schema = 1,
Type = "completion",
Description = "Gets the intent of the user.",
Completion =
{
Temperature = 0.1,
TopP = 0.5,
PresencePenalty = 0.0,
FrequencyPenalty = 0.0,
MaxTokens = 500
// StopSequences = null,
// ChatSystemPprompt = null;
},
Input =
{
Parameters = new List<PromptTemplateConfig.InputParameter>
{
new PromptTemplateConfig.InputParameter
{
Name="original_question",
Description="The user's request.",
DefaultValue=""
},
new PromptTemplateConfig.InputParameter
{
Name="search_results",
Description="Vector Search results from Azure Cognitive Search.",
DefaultValue=""
}
}
}
};
// Create the SemanticFunctionConfig object
PromptTemplate promptTemplate = new PromptTemplate(
sk_prompt,
promptConfig,
kernel
);
SemanticFunctionConfig functionConfig = new SemanticFunctionConfig(promptConfig, promptTemplate);
// Register the GetIntent function with the Kernel
ISKFunction getIntentFunction = kernel.RegisterSemanticFunction("CustomPlugin", "GetIntent", functionConfig);
// Configure a prompt as a plug in. We'll define the path to the plug in here and setup
// the plug in later.
var pluginsDirectory = Path.Combine(System.IO.Directory.GetCurrentDirectory(), "Plugins");
var customPlugin = kernel.ImportSemanticSkillFromDirectory(pluginsDirectory, "CustomPlugin");

Console.WriteLine("Semantic Function GetIntent with SK has been completed.");

Expand Down Expand Up @@ -211,7 +161,7 @@ app.MapPost("/completion", async ([FromServices] IKernel kernel, [FromBody] Comp
["search_results"] = stringBuilderResults.ToString()
};
// Use SK Chaining to Invoke Semantic Function
string completion = (await kernel.RunAsync(variables, getIntentFunction)).Result;
string completion = (await kernel.RunAsync(variables, customPlugin["GetIntent"])).Result;
Console.WriteLine(completion);

Console.WriteLine("Implementation of RAG using SK, C# and Azure Cognitive Search has been completed.");
Expand All @@ -235,21 +185,104 @@ public record CompletionRequest (string Question) {}
public record CompletionResponse (string Completion) {}
```

Last, but not least we changed the app.Run() to be async.
Note at the end of the last section of code, we replaced the app.Run() from the original code with the async version.

```csharp
await app.RunAsync();
```

Also, as we've replaced the `weatherforecast` function, we can remove the `WeatherForecast` record. So, you can delete the following lines which should be at the end of the `Program.cs` file.

```csharp
record WeatherForecast(DateOnly Date, int TemperatureC, string? Summary)
{
public int TemperatureF => 32 + (int)(TemperatureC / 0.5556);
}
```

### Create a Plug In

Next we will create a plug in that will be used to define the prompt that will be sent to the Azure OpenAI API.

In the root of the project create a folder called `Plugins` and in that folder create another folder called `CustomPlugin`.

```shell
mkdir Plugins
cd Plugins
mkdir CustomPlugin
```

The Plug In that we're going to create will be called `GetIntent`, so let's create a folder for that too.

```shell
cd CustomPlugin
mkdir GetIntent
```

Under the `GetIntent` folder we will create two files. One file will provide the template for the prompt that we want to use with Azure OpenAI. The other file will provide the configuration parameters.

Create a file called `skprompt.txt` and add the following text.

```text
Question: {{$original_question}}
Do not use any other data.
Only use the movie data below when responding.
{{$search_results}}
```

Next, create a file named `config.json` and add the following.

```json
{
"schema": 1,
"type": "completion",
"description": "Gets the intent of the user.",
"completion": {
"max_tokens": 500,
"temperature": 0.1,
"top_p": 0.5,
"presence_penalty": 0.0,
"frequency_penalty": 0.0
},
"input": {
"parameters": [
{
"name": "original_question",
"description": "The user's request.",
"defaultValue": ""
},
{
"name": "search_results",
"description": "Vector Search results from Azure Cognitive Search.",
"defaultValue": ""
}
]
}
}
```

When you've completed the above steps, your folder structure should look like this.

```text
acs-sk-csharp
├── Plugins
│ ├── CustomPlugin
│ │ ├── GetIntent
│ │ │ ├── config.json
│ │ │ ├── skprompt.txt
```

### Test the App

Now that we have all the code in place let'c compile and run it.
Now that we have all the code in place let's compile and run it.

```csharp
dotnet run
```

Once the app is started, open a browser and navigate to http://127.0.0.1:5291/swagger/index.html
Once the app is started, open a browser and navigate to http://127.0.0.1:5291/swagger/index.html
>**Note:** the port number may be different to `5291`, so double check the output from the `dotnet run` command.
Click on the "POST /completion" endpoint, click on "Try it out", enter a Prompt, "List the movies about ships on the water.", then click on "Execute".

Expand Down

0 comments on commit 4d6e461

Please sign in to comment.