MindWorkAI/AI-Studio
MindWorkAI/AI-Studio
1
0
Fork 0
mirror of https://github.com/MindWorkAI/AI-Studio.git synced 2025-04-28 18:59:47 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

Implemented prompting

Browse Source
This commit is contained in:
Thorsten Sommer 2024-12-31 19:34:42 +01:00
parent 36487b04a5
commit f8245c4fb8
Signed by: tsommer
GPG Key ID: 371BBA77A02C0108
2 changed files with 348 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
app/MindWork AI Studio/Assistants/ERI/AssistantERI.razor
View File

@ -313,9 +313,16 @@ else
Provider selection for generation Provider selection for generation
</MudText> </MudText>
<MudText Typo="Typo.body1" Class="mb-1"> <MudText Typo="Typo.body1" Class="mb-2">
The task of writing the ERI server for you is very complex. Therefore, a very powerful LLM is needed to successfully accomplish this task. The task of writing the ERI server for you is very complex. Therefore, a very powerful LLM is needed to successfully accomplish this task.
Small local models will probably not be sufficient. Instead, try using a large cloud-based or a large self-hosted model. Small local models will probably not be sufficient. Instead, try using a large cloud-based or a large self-hosted model.
</MudText> </MudText>
<MudText Typo="Typo.body1" Class="mb-2">
<b>Important:</b> The LLM may need to generate many files. This reaches the request limit of most providers. Typically, only a certain number
of requests can be made per minute, and only a maximum number of tokens can be generated per minute. AI Studio automatically considers this.
<b>However, generating all the files takes a certain amount of time.</b> Local or self-hosted models may work without these limitations
and can generate responses faster. AI Studio dynamically adapts its behavior and always tries to achieve the fastest possible data processing.
</MudText>
<ProviderSelection @bind-ProviderSettings="@this.providerSettings" ValidateProvider="@this.ValidatingProvider"/> <ProviderSelection @bind-ProviderSettings="@this.providerSettings" ValidateProvider="@this.ValidatingProvider"/>

341
app/MindWork AI Studio/Assistants/ERI/AssistantERI.razor.cs
View File

@ -1,3 +1,6 @@
using System.Text;
using System.Text.RegularExpressions;
using AIStudio.Chat; using AIStudio.Chat;
using AIStudio.Dialogs; using AIStudio.Dialogs;
using AIStudio.Settings.DataModel; using AIStudio.Settings.DataModel;
@ -29,13 +32,274 @@ public partial class AssistantERI : AssistantBaseCore
realize a Retrieval-Augmented Generation (RAG) process with external data. realize a Retrieval-Augmented Generation (RAG) process with external data.
"""; """;
protected override string SystemPrompt => protected override string SystemPrompt
$""" {
get
{
var sb = new StringBuilder();
"""; //
// ---------------------------------
// Introduction
// ---------------------------------
//
var programmingLanguage = this.selectedProgrammingLanguage is ProgrammingLanguages.OTHER ? this.otherProgrammingLanguage : this.selectedProgrammingLanguage.Name();
sb.Append($"""
# Introduction
You are an experienced {programmingLanguage} developer. Your task is to implement an API server in
the {programmingLanguage} language according to the following OpenAPI Description (OAD):
```
{this.eriSpecification}
```
The server realizes the data retrieval component for a "Retrieval-Augmentation Generation" (RAG) process.
The server is called "{this.serverName}" and is described as follows:
```
{this.serverDescription}
```
""");
//
// ---------------------------------
// Data Source
// ---------------------------------
//
sb.Append("""
# Data Source
""");
switch (this.selectedDataSource)
{
case DataSources.CUSTOM:
sb.Append($"""
The data source for the retrieval process is described as follows:
```
{this.otherDataSource}
```
""");
if(!string.IsNullOrWhiteSpace(this.dataSourceHostname))
{
sb.Append($"""
The data source is accessible via the hostname `{this.dataSourceHostname}`.
""");
}
if(this.dataSourcePort is not null)
{
sb.Append($"""
The data source is accessible via port `{this.dataSourcePort}`.
""");
}
break;
case DataSources.FILE_SYSTEM:
sb.Append("""
The data source for the retrieval process is the local file system. Use a placeholder for the data source path.
""");
break;
default:
case DataSources.OBJECT_STORAGE:
case DataSources.KEY_VALUE_STORE:
case DataSources.DOCUMENT_STORE:
case DataSources.RELATIONAL_DATABASE:
case DataSources.GRAPH_DATABASE:
sb.Append($"""
The data source for the retrieval process is an "{this.dataSourceProductName}" database running on the
host `{this.dataSourceHostname}` and is accessible via port `{this.dataSourcePort}`.
""");
break;
}
//
// ---------------------------------
// Authentication and Authorization
// ---------------------------------
//
sb.Append("""
# Authentication and Authorization
The process for authentication and authorization is two-step. Step 1: Users must authenticate
with the API server using a `POST` call on `/auth` with the chosen method. If this step is
successful, the API server returns a token. Step 2: This token is then required for all
other API calls.
Important notes:
- Calls to `/auth` and `/auth/methods` are accessible without authentication. All other API
endpoints require a valid step 2 token.
- It is possible that a token (step 1 token) is desired as `authMethod`. These step 1 tokens
must never be accepted as valid tokens for step 2.
- The OpenAPI Description (OAD) for `/auth` is not complete. This is because, at the time of
writing the OAD, it is not known what kind of authentication is desired. Therefore, you must
supplement the corresponding fields or data in the implementation. Example: If username/password
is desired, you must expect and read both. If both token and username/password are desired, you
must dynamically read the `authMethod` and expect and evaluate different fields accordingly.
The following authentications and authorizations should be implemented for the API server:
""");
foreach (var auth in this.selectedAuthenticationMethods)
sb.Append($"- {auth.ToPrompt()}");
if(this.IsUsingKerberos())
{
sb.Append($"""
The server will run on {this.selectedOperatingSystem.Name()} operating systems. Keep
this in mind when implementing the SSO with Kerberos.
""");
}
//
// ---------------------------------
// Security
// ---------------------------------
//
sb.Append($"""
# Security
The following security requirement for `allowedProviderType` was chosen: `{this.allowedLLMProviders}`
""");
//
// ---------------------------------
// Retrieval Processes
// ---------------------------------
//
sb.Append($"""
# Retrieval Processes
You are implementing the following data retrieval processes:
""");
var retrievalProcessCounter = 1;
foreach (var retrievalProcess in this.retrievalProcesses)
{
sb.Append($"""
## {retrievalProcessCounter++}. Retrieval Process
- Name: {retrievalProcess.Name}
- Description:
```
{retrievalProcess.Description}
```
""");
if(retrievalProcess.ParametersDescription?.Count > 0)
{
sb.Append("""
This retrieval process recognizes the following parameters:
""");
var parameterCounter = 1;
foreach (var (parameter, description) in retrievalProcess.ParametersDescription)
{
sb.Append($"""
- The {parameterCounter++} parameter is named "{parameter}":
```
{description}
```
""");
}
sb.Append("""
Please use sensible default values for the parameters. They are optional
for the user.
""");
}
if(retrievalProcess.Embeddings?.Count > 0)
{
sb.Append("""
The following embeddings are implemented for this retrieval process:
""");
var embeddingCounter = 1;
foreach (var embedding in retrievalProcess.Embeddings)
{
sb.Append($"""
- {embeddingCounter++}. Embedding
- Name: {embedding.EmbeddingName}
- Type: {embedding.EmbeddingType}
- Description:
```
{embedding.Description}
```
- When used:
```
{embedding.UsedWhen}
```
""");
}
}
}
//
// ---------------------------------
// Additional Libraries
// ---------------------------------
//
if (!string.IsNullOrWhiteSpace(this.additionalLibraries))
{
sb.Append($"""
# Additional Libraries
You use the following libraries for your implementation:
{this.additionalLibraries}
""");
}
//
// ---------------------------------
// Remarks
// ---------------------------------
//
sb.Append("""
# Remarks
- You do not ask follow-up questions.
- You consider the security of the implementation by applying the Security by Design principle.
- Your output is formatted as Markdown. Code is formatted as code blocks. For every file, you
create a separate code block with its file path and name as chapter title.
""");
return sb.ToString();
}
}
protected override IReadOnlyList<IButtonData> FooterButtons => []; protected override IReadOnlyList<IButtonData> FooterButtons => [];
protected override bool ShowEntireChatThread => true;
protected override string SubmitText => "Create the ERI server"; protected override string SubmitText => "Create the ERI server";
protected override Func<Task> SubmitAction => this.GenerateServer; protected override Func<Task> SubmitAction => this.GenerateServer;
@ -613,6 +877,25 @@ public partial class AssistantERI : AssistantBaseCore
await this.AutoSave(); await this.AutoSave();
} }
[GeneratedRegex("""
"([\\/._\w-]+)"
""", RegexOptions.NonBacktracking)]
private static partial Regex FileExtractRegex();
private IEnumerable<string> ExtractFiles(string fileListAnswer)
{
//
// We asked the LLM for answering using a specific JSON scheme.
// However, the LLM might not follow this scheme. Therefore, we
// need to parse the answer and extract the files.
// The parsing strategy is to look for all strings.
//
var matches = FileExtractRegex().Matches(fileListAnswer);
foreach (Match match in matches)
if(match.Groups[1].Value is {} file && !string.IsNullOrWhiteSpace(file) && !file.Equals("files", StringComparison.OrdinalIgnoreCase))
yield return file;
}
private async Task GenerateServer() private async Task GenerateServer()
{ {
if(this.IsNoneERIServerSelected) if(this.IsNoneERIServerSelected)
@ -635,5 +918,57 @@ public partial class AssistantERI : AssistantBaseCore
this.AddInputIssue("The ERI specification could not be loaded. Please try again later."); this.AddInputIssue("The ERI specification could not be loaded. Please try again later.");
return; return;
} }
this.CreateChatThread();
//
// ---------------------------------
// Ask for files (viewable in the chat)
// ---------------------------------
//
var time = this.AddUserRequest("""
Please list all the files you want to create. Provide the result as a Markdown list.
Start with a brief message that these are the files we are now creating.
""", true);
await this.AddAIResponseAsync(time);
//
// ---------------------------------
// Ask for files, again (JSON output, invisible)
// ---------------------------------
//
time = this.AddUserRequest("""
Please format all the files you want to create as a JSON object, without Markdown.
Use the following JSON schema:
{
[
"path/to/file1",
"path/to/file2"
]
}
""", true);
var fileListAnswer = await this.AddAIResponseAsync(time, true);
foreach (var file in this.ExtractFiles(fileListAnswer))
{
this.Logger.LogInformation($"The LLM want to create the file: '{file}'");
//
// ---------------------------------
// Ask the AI to create another file
// ---------------------------------
//
time = this.AddUserRequest($"""
Please create the file `{file}`. Your output is formatted in Markdown
using the following template:
## file/path
```language
content of the file
```
""", true);
await this.AddAIResponseAsync(time);
}
} }
} }