mirror of
https://github.com/MindWorkAI/AI-Studio.git
synced 2026-11-14 22:08:27 +00:00
Added the Model Comparison assistant for blind comparisons of two models (#1009)
Co-authored-by: Thorsten Sommer <SommerEngineering@users.noreply.github.com>
This commit is contained in:
1 parent
12f26445ba
commit
b7255694cd
221 files changed
+19989
-90
No files matched your search
@@ -0,0 +1,159 @@
|
||||
using AIStudio.Provider;
|
||||
using AIStudio.Settings;
|
||||
using AIStudio.Settings.DataModel;
|
||||
using AIStudio.Tools;
|
||||
using AIStudio.Tools.ToolCallingSystem;
|
||||
|
||||
namespace AIStudio.Tests.Tools;
|
||||
|
||||
/// <summary>
|
||||
/// Checks what a selection of tools or data sources offers when several models work with it.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// A rule which lets one model have more than another would make a comparison unfair, or send data to a model which
|
||||
/// may not see it. A rule which gives a single model less than before would change the chat.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class SharedSelectionRulesTests
|
||||
{
|
||||
private static readonly IDataSource ALPHA = DataSource("alpha");
|
||||
private static readonly IDataSource BETA = DataSource("beta");
|
||||
private static readonly IDataSource GAMMA = DataSource("gamma");
|
||||
|
||||
[Test]
|
||||
public void TheLowestConfidenceCounts()
|
||||
{
|
||||
Assert.That(SharedSelectionRules.LowestConfidence([ConfidenceLevel.HIGH, ConfidenceLevel.LOW, ConfidenceLevel.MEDIUM]), Is.EqualTo(ConfidenceLevel.LOW));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void WithoutAModelThereIsNoConfidence()
|
||||
{
|
||||
Assert.That(SharedSelectionRules.LowestConfidence([]), Is.EqualTo(ConfidenceLevel.NONE));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ToolsNeedEveryModelToCallThem()
|
||||
{
|
||||
var unavailable = new ToolCallingAvailability(false, "The second model cannot call tools.");
|
||||
|
||||
Assert.That(SharedSelectionRules.ToolCallingOfAll([ToolCallingAvailability.Available(), unavailable]), Is.EqualTo(unavailable), "The reason of the model which cannot call tools has to reach the user.");
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ToolsAreAvailableWhenEveryModelCallsThem()
|
||||
{
|
||||
Assert.That(SharedSelectionRules.ToolCallingOfAll([ToolCallingAvailability.Available(), ToolCallingAvailability.Available()]).IsAvailable, Is.True);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void SemanticSearchNeedsEveryModelToSearchItself()
|
||||
{
|
||||
var blocked = new EffectiveRetrievalMode(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.MODEL_CANNOT_USE_TOOLS);
|
||||
|
||||
var mode = SharedSelectionRules.RetrievalModeOfAll([new(DataSourceRetrievalMode.SEMANTIC_SEARCH, ToolOfferBlockReason.NONE), blocked]);
|
||||
|
||||
Assert.That(mode, Is.EqualTo(blocked), "One model searching itself while the other gets passages with every message would make the chats unequal.");
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void TheReasonComesAlongWhateverTheUserPrefers()
|
||||
{
|
||||
var blocked = new EffectiveRetrievalMode(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.PROVIDER_CONFIDENCE_TOO_LOW);
|
||||
|
||||
var mode = SharedSelectionRules.RetrievalModeOfAll([new(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.NONE), blocked]);
|
||||
|
||||
Assert.That(mode, Is.EqualTo(blocked), "The data source selection leaves out a choice which is none, which it can tell only by the reason.");
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void EveryModelSearchingItselfKeepsSemanticSearch()
|
||||
{
|
||||
var searchesItself = new EffectiveRetrievalMode(DataSourceRetrievalMode.SEMANTIC_SEARCH, ToolOfferBlockReason.NONE);
|
||||
|
||||
Assert.That(SharedSelectionRules.RetrievalModeOfAll([searchesItself, searchesItself]), Is.EqualTo(searchesItself));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ThePreferenceOfTheUserStays()
|
||||
{
|
||||
var everyMessage = new EffectiveRetrievalMode(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.NONE);
|
||||
|
||||
Assert.That(SharedSelectionRules.RetrievalModeOfAll([everyMessage, everyMessage]), Is.EqualTo(everyMessage));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ASingleModelKeepsItsOwnMode()
|
||||
{
|
||||
foreach (var mode in new EffectiveRetrievalMode[]
|
||||
{
|
||||
new(DataSourceRetrievalMode.SEMANTIC_SEARCH, ToolOfferBlockReason.NONE),
|
||||
new(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.NONE),
|
||||
new(DataSourceRetrievalMode.EVERY_MESSAGE, ToolOfferBlockReason.TOOLS_SWITCHED_OFF),
|
||||
})
|
||||
Assert.That(SharedSelectionRules.RetrievalModeOfAll([mode]), Is.EqualTo(mode));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void OnlyTheDataSourcesEveryModelMayUseRemain()
|
||||
{
|
||||
var first = new AllowedSelectedDataSources([ALPHA, BETA, GAMMA], [ALPHA, BETA], [], [], []);
|
||||
var second = new AllowedSelectedDataSources([GAMMA, ALPHA], [ALPHA], [], [], []);
|
||||
|
||||
var shared = SharedSelectionRules.DataSourcesOfAll([first, second]);
|
||||
|
||||
using (Assert.EnterMultipleScope())
|
||||
{
|
||||
Assert.That(shared.AllowedDataSources, Is.EqualTo(new[] { ALPHA, GAMMA }), "The order of the first model counts.");
|
||||
Assert.That(shared.SelectedDataSources, Is.EqualTo(new[] { ALPHA }));
|
||||
}
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AWaitingDataSourceStaysWaitingOnlyWhenEveryModelMayUseIt()
|
||||
{
|
||||
var first = new AllowedSelectedDataSources([], [], [ALPHA, BETA], [GAMMA], [BETA]);
|
||||
var second = new AllowedSelectedDataSources([], [], [ALPHA], [GAMMA], []);
|
||||
|
||||
var shared = SharedSelectionRules.DataSourcesOfAll([first, second]);
|
||||
|
||||
using (Assert.EnterMultipleScope())
|
||||
{
|
||||
Assert.That(shared.DataSourcesAwaitingReindex, Is.EqualTo(new[] { ALPHA }));
|
||||
Assert.That(shared.DataSourcesNeedingRepair, Is.EqualTo(new[] { GAMMA }));
|
||||
Assert.That(shared.UnreachableDataSources, Is.Empty);
|
||||
}
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ASingleModelKeepsItsDataSources()
|
||||
{
|
||||
var only = new AllowedSelectedDataSources([ALPHA, BETA], [BETA], [GAMMA], [], [ALPHA]);
|
||||
|
||||
var shared = SharedSelectionRules.DataSourcesOfAll([only]);
|
||||
|
||||
using (Assert.EnterMultipleScope())
|
||||
{
|
||||
Assert.That(shared.AllowedDataSources, Is.EqualTo(only.AllowedDataSources));
|
||||
Assert.That(shared.SelectedDataSources, Is.EqualTo(only.SelectedDataSources));
|
||||
Assert.That(shared.DataSourcesAwaitingReindex, Is.EqualTo(only.DataSourcesAwaitingReindex));
|
||||
Assert.That(shared.DataSourcesNeedingRepair, Is.EqualTo(only.DataSourcesNeedingRepair));
|
||||
Assert.That(shared.UnreachableDataSources, Is.EqualTo(only.UnreachableDataSources));
|
||||
}
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void WithoutAModelThereIsNoDataSource()
|
||||
{
|
||||
Assert.That(SharedSelectionRules.DataSourcesOfAll([]).AllowedDataSources, Is.Empty);
|
||||
}
|
||||
|
||||
private static IDataSource DataSource(string id) => new DataSourceLocalFile
|
||||
{
|
||||
Num = 1,
|
||||
Id = id,
|
||||
Name = id,
|
||||
Type = DataSourceType.LOCAL_FILE,
|
||||
ConfidenceLevel = ConfidenceLevel.LOW,
|
||||
};
|
||||
}
|
||||
Reference in new issue
Block a user