mirror of
https://github.com/MindWorkAI/AI-Studio.git
synced 2026-10-05 04:49:40 +00:00
Added the exact token count where the provider reports it (#989)
Build and Release / Verify (push) Waiting to run
Build and Release / Determine run mode (push) Waiting to run
Build and Release / Read metadata (push) Blocked by required conditions
Build and Release / Sync Flatpak repo (push) Blocked by required conditions
Build and Release / Collect Flatpak artifacts (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-apple-darwin, osx-arm64, macos-latest, aarch64-apple-darwin, dmg,app,updater, dmg) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-pc-windows-msvc.exe, win-arm64, windows-latest, aarch64-pc-windows-msvc, nsis,updater, nsis) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-unknown-linux-gnu, linux-arm64, ubuntu-22.04-arm, aarch64-unknown-linux-gnu, appimage,updater, appimage) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-apple-darwin, osx-x64, macos-latest, x86_64-apple-darwin, dmg,app,updater, dmg) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-pc-windows-msvc.exe, win-x64, windows-latest, x86_64-pc-windows-msvc, nsis,updater, nsis) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-unknown-linux-gnu, linux-x64, ubuntu-22.04, x86_64-unknown-linux-gnu, appimage,updater, appimage) (push) Blocked by required conditions
Build and Release / Prepare & create release (push) Blocked by required conditions
Build and Release / Publish release (push) Blocked by required conditions
Build and Release / Verify (push) Waiting to run
Build and Release / Determine run mode (push) Waiting to run
Build and Release / Read metadata (push) Blocked by required conditions
Build and Release / Sync Flatpak repo (push) Blocked by required conditions
Build and Release / Collect Flatpak artifacts (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-apple-darwin, osx-arm64, macos-latest, aarch64-apple-darwin, dmg,app,updater, dmg) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-pc-windows-msvc.exe, win-arm64, windows-latest, aarch64-pc-windows-msvc, nsis,updater, nsis) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-aarch64-unknown-linux-gnu, linux-arm64, ubuntu-22.04-arm, aarch64-unknown-linux-gnu, appimage,updater, appimage) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-apple-darwin, osx-x64, macos-latest, x86_64-apple-darwin, dmg,app,updater, dmg) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-pc-windows-msvc.exe, win-x64, windows-latest, x86_64-pc-windows-msvc, nsis,updater, nsis) (push) Blocked by required conditions
Build and Release / Build app (${{ matrix.dotnet_runtime }}) (-x86_64-unknown-linux-gnu, linux-x64, ubuntu-22.04, x86_64-unknown-linux-gnu, appimage,updater, appimage) (push) Blocked by required conditions
Build and Release / Prepare & create release (push) Blocked by required conditions
Build and Release / Publish release (push) Blocked by required conditions
Co-authored-by: Thorsten Sommer <SommerEngineering@users.noreply.github.com>
This commit is contained in:
1 parent
82986afe62
commit
be1e6532fb
34 files changed
+1360
-104
No files matched your search
@@ -0,0 +1,184 @@
|
||||
using AIStudio.Chat;
|
||||
using AIStudio.Provider;
|
||||
|
||||
namespace AIStudio.Tests.Chat;
|
||||
|
||||
/// <summary>
|
||||
/// Checks when what a provider reported for a conversation still describes it.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// A reported number is shown as exact, so a wrong one does more harm than no number at all: it
|
||||
/// tells a person that their chat fits while it does not. Every case below is one where the thread
|
||||
/// moved on after the report, and the estimate has to take over -- or one where it came back to
|
||||
/// exactly the state the report was taken in, and the report counts again.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class ChatThreadReportedHistoryTests
|
||||
{
|
||||
private static readonly DateTimeOffset START = new(2026, 9, 23, 10, 0, 0, TimeSpan.Zero);
|
||||
|
||||
private static readonly Model MODEL = new("model-a", null);
|
||||
|
||||
private static readonly Model OTHER_MODEL = new("model-b", null);
|
||||
|
||||
[Test]
|
||||
public void TheReportOfTheLastAnswerCounts()
|
||||
{
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2));
|
||||
|
||||
var history = thread.ReportedHistoryFor(MODEL);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(history.IsKnown, Is.True);
|
||||
Assert.That(history.PromptTokens, Is.EqualTo(1200));
|
||||
|
||||
//
|
||||
// The answer comes along as text rather than as the provider's number for it, which
|
||||
// would include the reasoning the next request never carries:
|
||||
//
|
||||
Assert.That(history.LastAnswer, Is.EqualTo("Answer 2"));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AnAnswerWithoutTextAddsNothingToThePrompt()
|
||||
{
|
||||
//
|
||||
// An answer which consists of reasoning only. It may be kept to be read, but without any
|
||||
// text it is never sent, so what the provider counted is all the next request carries of
|
||||
// the conversation so far.
|
||||
//
|
||||
var reasoningOnly = new ContentText { Text = string.Empty };
|
||||
reasoningOnly.RecordReportedUsage(TokenUsage.Of(1200), MODEL.Id, 2);
|
||||
var thread = Thread(Question(1), Block(ChatRole.AI, reasoningOnly, 2));
|
||||
|
||||
var history = thread.ReportedHistoryFor(MODEL);
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(history.IsKnown, Is.True);
|
||||
Assert.That(history.PromptTokens, Is.EqualTo(1200));
|
||||
Assert.That(history.LastAnswer, Is.Empty);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AMessageWhoseRequestFailedIsNotLeftOut()
|
||||
{
|
||||
//
|
||||
// A request which was turned down leaves the question standing and removes the empty
|
||||
// answer. The report before it knows nothing about that question -- which may well be the
|
||||
// very message that made the chat too large.
|
||||
//
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2), Question(3));
|
||||
|
||||
Assert.That(thread.ReportedHistoryFor(MODEL).IsKnown, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AnAnswerBeingWrittenDoesNotBorrowTheReportBeforeIt()
|
||||
{
|
||||
var streaming = Block(ChatRole.AI, new ContentText { Text = "Half an ans", IsStreaming = true }, 4);
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2), Question(3), streaming);
|
||||
|
||||
Assert.That(thread.ReportedHistoryFor(MODEL).IsKnown, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AnAnswerWithoutAReportDoesNotBorrowTheReportBeforeIt()
|
||||
{
|
||||
//
|
||||
// What an answer looks like whose provider reports nothing, or whose API is not read for a
|
||||
// report yet: finished, but without a report of its own.
|
||||
//
|
||||
var withoutReport = Block(ChatRole.AI, new ContentText { Text = "Second answer" }, 4);
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2), Question(3), withoutReport);
|
||||
|
||||
Assert.That(thread.ReportedHistoryFor(MODEL).IsKnown, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void DeletingAnEarlierMessageOutdatesTheReport()
|
||||
{
|
||||
//
|
||||
// Deleting a large message to make room is exactly when somebody watches the number, and
|
||||
// the deleted message is still inside what the provider reported.
|
||||
//
|
||||
var firstQuestion = Question(1);
|
||||
var thread = Thread(firstQuestion, Answer(2, promptTokens: 1200, blockCount: 2), Question(3), Answer(4, promptTokens: 2600, blockCount: 4));
|
||||
|
||||
thread.Remove(firstQuestion.Content!);
|
||||
|
||||
Assert.That(thread.ReportedHistoryFor(MODEL).IsKnown, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AnotherModelOutdatesTheReport()
|
||||
{
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2));
|
||||
|
||||
Assert.That(thread.ReportedHistoryFor(OTHER_MODEL).IsKnown, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void EditingTheLastMessageBringsTheReportBeforeItBack()
|
||||
{
|
||||
var lastQuestion = Question(3);
|
||||
var lastAnswer = Answer(4, promptTokens: 2600, blockCount: 4);
|
||||
var thread = Thread(Question(1), Answer(2, promptTokens: 1200, blockCount: 2), lastQuestion, lastAnswer);
|
||||
|
||||
//
|
||||
// What editing the last message does to the thread: the question goes back into the
|
||||
// composer, and its answer goes with it.
|
||||
//
|
||||
thread.Remove(lastQuestion.Content!);
|
||||
thread.Remove(lastAnswer.Content!);
|
||||
|
||||
var history = thread.ReportedHistoryFor(MODEL);
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(history.IsKnown, Is.True, "The first answer is the last block again, with exactly the blocks it was reported for.");
|
||||
Assert.That(history.PromptTokens, Is.EqualTo(1200));
|
||||
Assert.That(history.LastAnswer, Is.EqualTo("Answer 2"));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void RollingBackToAnAnswerBringsItsReportBack()
|
||||
{
|
||||
var firstAnswer = Answer(2, promptTokens: 1200, blockCount: 2);
|
||||
var thread = Thread(Question(1), firstAnswer, Question(3), Answer(4, promptTokens: 2600, blockCount: 4));
|
||||
|
||||
thread.RollBackTo(firstAnswer.Content!);
|
||||
|
||||
var history = thread.ReportedHistoryFor(MODEL);
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(history.IsKnown, Is.True);
|
||||
Assert.That(history.PromptTokens, Is.EqualTo(1200));
|
||||
Assert.That(history.LastAnswer, Is.EqualTo("Answer 2"));
|
||||
});
|
||||
}
|
||||
|
||||
private static ChatThread Thread(params ContentBlock[] blocks) => new()
|
||||
{
|
||||
Blocks = [..blocks],
|
||||
};
|
||||
|
||||
private static ContentBlock Question(int minute) => Block(ChatRole.USER, new ContentText { Text = $"Question {minute}" }, minute);
|
||||
|
||||
private static ContentBlock Answer(int minute, int promptTokens, int blockCount)
|
||||
{
|
||||
var answer = new ContentText { Text = $"Answer {minute}" };
|
||||
answer.RecordReportedUsage(TokenUsage.Of(promptTokens), MODEL.Id, blockCount);
|
||||
return Block(ChatRole.AI, answer, minute);
|
||||
}
|
||||
|
||||
private static ContentBlock Block(ChatRole role, ContentText content, int minute) => new()
|
||||
{
|
||||
Time = START.AddMinutes(minute),
|
||||
ContentType = ContentType.TEXT,
|
||||
Content = content,
|
||||
Role = role,
|
||||
};
|
||||
}
|
||||
@@ -56,7 +56,8 @@ public sealed class ConversationPartsTests
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Texts, Is.EqualTo(new[] { "You are helpful.", "What is the capital of France?", "Paris." }));
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "And of Italy?" }));
|
||||
Assert.That(parts.DraftText, Is.EqualTo("And of Italy?"));
|
||||
Assert.That(parts.GrowingTexts, Is.Empty, "The draft is a part of its own, and it stands nowhere else.");
|
||||
});
|
||||
}
|
||||
|
||||
@@ -77,7 +78,8 @@ public sealed class ConversationPartsTests
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Texts, Is.EqualTo(new[] { "A question." }));
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "The answer so far", "a draft" }));
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "The answer so far" }));
|
||||
Assert.That(parts.DraftText, Is.EqualTo("a draft"));
|
||||
});
|
||||
}
|
||||
|
||||
@@ -135,7 +137,7 @@ public sealed class ConversationPartsTests
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Texts, Is.Empty);
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "Hello" }));
|
||||
Assert.That(parts.DraftText, Is.EqualTo("Hello"));
|
||||
});
|
||||
}
|
||||
|
||||
@@ -149,6 +151,7 @@ public sealed class ConversationPartsTests
|
||||
{
|
||||
Assert.That(parts.Texts, Is.Empty);
|
||||
Assert.That(parts.GrowingTexts, Is.Empty);
|
||||
Assert.That(parts.DraftText, Is.Empty);
|
||||
});
|
||||
}
|
||||
|
||||
@@ -188,7 +191,11 @@ public sealed class ConversationPartsTests
|
||||
|
||||
var parts = ConversationParts.Of(new() { Blocks = [block] }, string.Empty, "And this one.", [FileAttachment.FromPath(draft)], imagesAreSent: true, toolDefinitions: null);
|
||||
|
||||
Assert.That(parts.Documents.Select(document => document.FileName), Is.EqualTo(new[] { "older.txt", "draft.txt" }));
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Documents.Select(document => document.FileName), Is.EqualTo(new[] { "older.txt" }));
|
||||
Assert.That(parts.DraftDocuments.Select(document => document.FileName), Is.EqualTo(new[] { "draft.txt" }), "Both count, each in the part it belongs to.");
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
@@ -201,7 +208,7 @@ public sealed class ConversationPartsTests
|
||||
|
||||
var parts = ConversationParts.Of(null, string.Empty, "Here", [attachment], imagesAreSent: true, toolDefinitions: null);
|
||||
|
||||
Assert.That(parts.Documents, Is.Empty);
|
||||
Assert.That(parts.DraftDocuments, Is.Empty);
|
||||
}
|
||||
|
||||
[Test]
|
||||
@@ -214,8 +221,29 @@ public sealed class ConversationPartsTests
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Documents.Select(entry => entry.FileName), Is.EqualTo(new[] { "notes.txt" }));
|
||||
Assert.That(parts.DraftDocuments.Select(entry => entry.FileName), Is.EqualTo(new[] { "notes.txt" }));
|
||||
Assert.That(parts.DraftImages, Is.EqualTo(1));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ImagesOfTheConversationAndOfTheDraftAreKeptApart()
|
||||
{
|
||||
//
|
||||
// A provider which reported the conversation so far counted the pictures in it as well,
|
||||
// but never the ones which are still waiting in the composer.
|
||||
//
|
||||
var sent = this.WriteFile("sent.png", "not really a png");
|
||||
var waiting = this.WriteFile("waiting.png", "not really a png either");
|
||||
var block = Block("Look at this.");
|
||||
((ContentText)block.Content!).FileAttachments.Add(FileAttachment.FromPath(sent));
|
||||
|
||||
var parts = ConversationParts.Of(new() { Blocks = [block] }, string.Empty, "And at this.", [FileAttachment.FromPath(waiting)], imagesAreSent: true, toolDefinitions: null);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Images, Is.EqualTo(1));
|
||||
Assert.That(parts.DraftImages, Is.EqualTo(1));
|
||||
});
|
||||
}
|
||||
|
||||
@@ -230,7 +258,7 @@ public sealed class ConversationPartsTests
|
||||
|
||||
var parts = ConversationParts.Of(null, string.Empty, "Look", [FileAttachment.FromPath(image)], imagesAreSent: false, toolDefinitions: null);
|
||||
|
||||
Assert.That(parts.Images, Is.Zero);
|
||||
Assert.That(parts.DraftImages, Is.Zero);
|
||||
}
|
||||
|
||||
[Test]
|
||||
@@ -250,7 +278,8 @@ public sealed class ConversationPartsTests
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Texts, Is.Empty);
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "What the web search found.", "What the page said." }));
|
||||
Assert.That(parts.ToolConversation, Is.EqualTo(new[] { "What the web search found.", "What the page said." }));
|
||||
Assert.That(parts.GrowingTexts, Is.Empty, "The tool conversation is a part of its own, so that its share can be named.");
|
||||
});
|
||||
}
|
||||
|
||||
@@ -267,7 +296,7 @@ public sealed class ConversationPartsTests
|
||||
|
||||
var parts = ConversationParts.Of(new() { Blocks = [running] }, string.Empty, string.Empty, null, imagesAreSent: true, toolDefinitions: null);
|
||||
|
||||
Assert.That(parts.GrowingTexts, Is.EqualTo(new[] { "The same page.", "The same page." }));
|
||||
Assert.That(parts.ToolConversation, Is.EqualTo(new[] { "The same page.", "The same page." }));
|
||||
}
|
||||
|
||||
[Test]
|
||||
@@ -288,7 +317,7 @@ public sealed class ConversationPartsTests
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(parts.Texts, Is.EqualTo(new[] { "Here is what I found." }));
|
||||
Assert.That(parts.GrowingTexts, Is.Empty);
|
||||
Assert.That(parts.ToolConversation, Is.Empty);
|
||||
});
|
||||
}
|
||||
|
||||
|
||||
@@ -25,7 +25,7 @@ public sealed class ConversationTokensTests
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
UncountedImages = images,
|
||||
Images = images,
|
||||
ImageLimits = new ImageLimits(null, allowed),
|
||||
};
|
||||
|
||||
@@ -42,7 +42,7 @@ public sealed class ConversationTokensTests
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
UncountedImages = 500,
|
||||
Images = 500,
|
||||
ImageLimits = ImageLimits.UNKNOWN,
|
||||
};
|
||||
|
||||
@@ -60,7 +60,7 @@ public sealed class ConversationTokensTests
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
UncountedImages = 20,
|
||||
Images = 20,
|
||||
ImageLimits = new ImageLimits(8, 100),
|
||||
};
|
||||
|
||||
@@ -77,7 +77,7 @@ public sealed class ConversationTokensTests
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
UncountedImages = 0,
|
||||
Images = 0,
|
||||
ImageLimits = new ImageLimits(null, 0),
|
||||
};
|
||||
|
||||
@@ -93,4 +93,64 @@ public sealed class ConversationTokensTests
|
||||
//
|
||||
Assert.That(ConversationTokens.UNAVAILABLE.TooManyImages, Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void PicturesAProviderCountedAreNotCalledUncounted()
|
||||
{
|
||||
//
|
||||
// What the provider reported for the conversation so far includes its pictures, however
|
||||
// it charges them. Only those still waiting in the composer are left for nobody to count.
|
||||
//
|
||||
var reported = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
HistoryIsReported = true,
|
||||
Images = 3,
|
||||
DraftImages = 1,
|
||||
};
|
||||
|
||||
var estimated = reported with { HistoryIsReported = false };
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(reported.UncountedImages, Is.EqualTo(1), "The provider counted the two which were sent.");
|
||||
Assert.That(estimated.UncountedImages, Is.EqualTo(3), "Without a report, nobody counted any of them.");
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void TooManyPicturesStaysTooManyWhenTheProviderCountedThem()
|
||||
{
|
||||
//
|
||||
// The limit is on how many pictures travel, not on what they cost. A provider which has
|
||||
// counted them still refuses the request which carries one too many.
|
||||
//
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
HistoryIsReported = true,
|
||||
Images = 10,
|
||||
ImageLimits = new ImageLimits(8, null),
|
||||
};
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(counted.UncountedImages, Is.Zero);
|
||||
Assert.That(counted.TooManyImages, Is.True);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void TheWholeNumberIsTheConversationAndTheDraft()
|
||||
{
|
||||
var counted = new ConversationTokens
|
||||
{
|
||||
IsKnown = true,
|
||||
HistoryTokens = 12_400,
|
||||
ToolTokens = 9_000,
|
||||
DraftTokens = 340,
|
||||
};
|
||||
|
||||
Assert.That(counted.Tokens, Is.EqualTo(12_740), "The tools' share is part of the conversation, not added on top of it.");
|
||||
}
|
||||
}
|
||||
@@ -51,4 +51,4 @@ public sealed class TokenAmountTests
|
||||
//
|
||||
Assert.That(TokenAmount.Format(tokens, GERMAN), Is.EqualTo(wanted));
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,129 @@
|
||||
using System.Text.Json;
|
||||
|
||||
using AIStudio.Provider;
|
||||
using AIStudio.Provider.OpenAI;
|
||||
|
||||
namespace AIStudio.Tests.Provider;
|
||||
|
||||
/// <summary>
|
||||
/// Checks that what a provider says a request cost is read off the stream, and asked for.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Both halves matter and neither is visible from the other: an OpenAI-compatible provider says
|
||||
/// nothing about the cost of a streamed request unless the request asks for it, and the line it
|
||||
/// then sends carries no content, so the reading side has to look for it apart from the text.
|
||||
/// Get either half wrong and the app silently keeps estimating, which looks exactly like a
|
||||
/// provider which reports nothing.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class ChatCompletionUsageTests
|
||||
{
|
||||
/// <summary>
|
||||
/// The last line of a streamed answer at a provider which was asked for the usage.
|
||||
/// </summary>
|
||||
private const string USAGE_LINE =
|
||||
"""
|
||||
{"id":"chatcmpl-1","object":"chat.completion.chunk","created":1,"model":"gpt-5","choices":[],"usage":{"prompt_tokens":1200,"completion_tokens":345,"total_tokens":1545}}
|
||||
""";
|
||||
|
||||
/// <summary>
|
||||
/// An ordinary line carrying a piece of the answer.
|
||||
/// </summary>
|
||||
private const string CONTENT_LINE =
|
||||
"""
|
||||
{"id":"chatcmpl-1","object":"chat.completion.chunk","created":1,"model":"gpt-5","choices":[{"index":0,"delta":{"content":"Hi"}}]}
|
||||
""";
|
||||
|
||||
[Test]
|
||||
public void TheFinalLineStatesWhatTheRequestCost()
|
||||
{
|
||||
var line = JsonSerializer.Deserialize<ChatCompletionDeltaStreamLine>(USAGE_LINE, ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(line!.GetUsage().IsKnown, Is.True);
|
||||
Assert.That(line.GetUsage().PromptTokens, Is.EqualTo(1200));
|
||||
|
||||
//
|
||||
// The line which carries the usage carries no answer, which is why it has to be read
|
||||
// before the content check drops it:
|
||||
//
|
||||
Assert.That(line.ContainsContent(), Is.False);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ALineOfTheAnswerStatesNoCost()
|
||||
{
|
||||
var line = JsonSerializer.Deserialize<ChatCompletionDeltaStreamLine>(CONTENT_LINE, ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(line!.GetUsage().IsKnown, Is.False);
|
||||
Assert.That(line.ContainsContent(), Is.True);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AStreamedRequestAsksForTheUsage()
|
||||
{
|
||||
var request = new ChatCompletionAPIRequest("gpt-5", [], true);
|
||||
var json = JsonSerializer.Serialize(request, ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.That(json, Does.Contain("""
|
||||
"stream_options":{"include_usage":true}
|
||||
"""));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ARequestWhichIsNotStreamedDoesNot()
|
||||
{
|
||||
var request = new ChatCompletionAPIRequest("gpt-5", [], false);
|
||||
var json = JsonSerializer.Serialize(request, ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.That(json, Does.Not.Contain("stream_options"));
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// A provider which sends the block but fills in nothing usable states nothing.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Several OpenAI-compatible servers send an empty or zeroed usage block on every line while
|
||||
/// streaming and the real numbers only at the end. Reading a zero as a fact would replace an
|
||||
/// estimate with a statement that the conversation costs nothing.
|
||||
/// </remarks>
|
||||
[Test]
|
||||
public void AnEmptyUsageBlockStatesNothing()
|
||||
{
|
||||
var line = JsonSerializer.Deserialize<ChatCompletionDeltaStreamLine>(
|
||||
"""
|
||||
{"id":"chatcmpl-1","object":"chat.completion.chunk","created":1,"model":"gpt-5","choices":[],"usage":{"prompt_tokens":0,"completion_tokens":0}}
|
||||
""", ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.That(line!.GetUsage().IsKnown, Is.False);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// The line a real LM Studio server sends, taken off the wire.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// It carries far more than we read, and it shows why the prompt is all we take: 102 of the 116
|
||||
/// completion tokens are the model's reasoning, which the next request never carries. Counting
|
||||
/// the completion would have put the history at 133 tokens, when what travels on is the prompt
|
||||
/// and an answer of a handful of tokens.
|
||||
/// </remarks>
|
||||
[Test]
|
||||
public void ARealServerLineIsRead()
|
||||
{
|
||||
var line = JsonSerializer.Deserialize<ChatCompletionDeltaStreamLine>(
|
||||
"""
|
||||
{"id":"chatcmpl-xb8mn282eff3tiu46xz8t3","object":"chat.completion.chunk","created":1789917744,"model":"google/gemma-4-12b-qat","system_fingerprint":"google/gemma-4-12b-qat","choices":[],"usage":{"prompt_tokens":17,"completion_tokens":116,"total_tokens":133,"completion_tokens_details":{"reasoning_tokens":102}}}
|
||||
""", ProviderJsonOptions.OPTIONS);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(line!.GetUsage().IsKnown, Is.True);
|
||||
Assert.That(line.GetUsage().PromptTokens, Is.EqualTo(17));
|
||||
});
|
||||
}
|
||||
}
|
||||
@@ -183,7 +183,38 @@ public sealed class ChatCompletionToolCallAccumulatorTests
|
||||
|
||||
Assert.That(part.Sources.Select(x => x.URL), Is.EqualTo(new[] { "https://example.org/" }), "Whatever the provider announced on that line reaches the user with it.");
|
||||
}
|
||||
|
||||
|
||||
[Test]
|
||||
public void TheLineWithoutChoicesStatesWhatTheRequestCost()
|
||||
{
|
||||
var accumulator = new ChatCompletionToolCallAccumulator();
|
||||
var part = accumulator.Process(Event("""{"choices":[],"usage":{"prompt_tokens":1200,"completion_tokens":345,"total_tokens":1545}}"""));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(part.Usage.IsKnown, Is.True, "The last line of the stream has no choices, and it must not be dropped for that.");
|
||||
Assert.That(part.Usage.PromptTokens, Is.EqualTo(1200));
|
||||
Assert.That(part.HasContent, Is.False, "It has nothing to show, though.");
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AUsageNextToTheLastTextIsReadAsWell()
|
||||
{
|
||||
//
|
||||
// Some providers put the usage on the line which carries the last piece of the answer
|
||||
// rather than on a line of its own.
|
||||
//
|
||||
var accumulator = new ChatCompletionToolCallAccumulator();
|
||||
var part = accumulator.Process(Event("""{"choices":[{"index":0,"delta":{"content":"Bye"}}],"usage":{"prompt_tokens":1200,"completion_tokens":2}}"""));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(part.TextDelta, Is.EqualTo("Bye"));
|
||||
Assert.That(part.Usage.PromptTokens, Is.EqualTo(1200));
|
||||
});
|
||||
}
|
||||
|
||||
private static ChatCompletionResponseMessage? Read(params string[] data)
|
||||
{
|
||||
var accumulator = new ChatCompletionToolCallAccumulator();
|
||||
|
||||
@@ -0,0 +1,169 @@
|
||||
using System.Runtime.CompilerServices;
|
||||
using System.Text.Json;
|
||||
|
||||
using AIStudio.Provider;
|
||||
using AIStudio.Provider.OpenAI;
|
||||
|
||||
using Microsoft.Extensions.Logging.Abstractions;
|
||||
|
||||
namespace AIStudio.Tests.Provider.ToolCalling;
|
||||
|
||||
/// <summary>
|
||||
/// Checks what a round of a tool calling conversation asks for, and what it passes on.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Every round of a tool conversation is a request of its own, and every one of them reports what
|
||||
/// it cost. Only the first one describes what the next question will be sent after: every later
|
||||
/// round carries the tool calls and their results on top, none of which is sent again once the
|
||||
/// answer stands. Passing on the last report instead would put the chat at the size of everything
|
||||
/// the tools returned, which is the one number a person watching their context window must not
|
||||
/// see as exact.
|
||||
///
|
||||
/// What a round asks for is one tool call at a time, wherever the provider lets it ask: a provider
|
||||
/// which rejects the question fails the whole request, so it is not asked at all.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class ChatCompletionToolCallingAdapterTests
|
||||
{
|
||||
private const string FIRST_ROUND_USAGE = """{"choices":[],"usage":{"prompt_tokens":1200,"completion_tokens":20}}""";
|
||||
|
||||
private const string SECOND_ROUND_USAGE = """{"choices":[],"usage":{"prompt_tokens":9800,"completion_tokens":150}}""";
|
||||
|
||||
[Test]
|
||||
public async Task OnlyTheFirstRoundPassesOnWhatItsRequestCost()
|
||||
{
|
||||
var adapter = Adapter(
|
||||
[
|
||||
"""{"choices":[{"index":0,"delta":{"tool_calls":[{"index":0,"id":"call_1","type":"function","function":{"name":"web_search","arguments":"{\"query\":\"weather\"}"}}]}}]}""",
|
||||
FIRST_ROUND_USAGE,
|
||||
"[DONE]",
|
||||
],
|
||||
[
|
||||
"""{"choices":[{"index":0,"delta":{"content":"It is sunny."}}]}""",
|
||||
SECOND_ROUND_USAGE,
|
||||
"[DONE]",
|
||||
]);
|
||||
|
||||
var firstRound = await Usages(adapter);
|
||||
|
||||
//
|
||||
// What the loop does between two rounds: the model's turn and the tool's result become part
|
||||
// of the next request.
|
||||
//
|
||||
adapter.RecordAssistantTurn();
|
||||
adapter.RecordToolResult("call_1", "Sunny, 24 degrees.");
|
||||
|
||||
var secondRound = await Usages(adapter);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(firstRound, Is.EqualTo(new[] { 1200 }), "The first round's prompt is the conversation up to the question.");
|
||||
Assert.That(secondRound, Is.Empty, "The second round's prompt holds the tool result as well, which the next question is not sent with.");
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public async Task ARoundWithoutToolCallsPassesItOnAsWell()
|
||||
{
|
||||
//
|
||||
// Offering tools does not mean the model uses them. Then the first round is the only one,
|
||||
// and its report is as good as the one of a request which offered none.
|
||||
//
|
||||
var adapter = Adapter(
|
||||
[
|
||||
"""{"choices":[{"index":0,"delta":{"content":"Hello."}}]}""",
|
||||
FIRST_ROUND_USAGE,
|
||||
"[DONE]",
|
||||
]);
|
||||
|
||||
Assert.That(await Usages(adapter), Is.EqualTo(new[] { 1200 }));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public async Task ARoundWhichOffersToolsAsksForOneCallAtATime()
|
||||
{
|
||||
Assert.That(await SentRequest(mayAskForSequentialToolCalls: true, includeTools: true), Does.Contain("\"parallel_tool_calls\":false"));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public async Task AProviderWhichRejectsTheQuestionIsNotAskedIt()
|
||||
{
|
||||
//
|
||||
// Hugging Face answers the question with a bad request. Its models may then ask for several
|
||||
// calls at once, which the loop works through one by one anyway.
|
||||
//
|
||||
Assert.That(await SentRequest(mayAskForSequentialToolCalls: false, includeTools: true), Does.Not.Contain("parallel_tool_calls"));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public async Task ARoundWithoutToolsDoesNotAskAboutToolCalls()
|
||||
{
|
||||
Assert.That(await SentRequest(mayAskForSequentialToolCalls: true, includeTools: false), Does.Not.Contain("parallel_tool_calls"));
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Runs one round and returns the request it sent, as it goes over the wire.
|
||||
/// </summary>
|
||||
private static async Task<string> SentRequest(bool mayAskForSequentialToolCalls, bool includeTools)
|
||||
{
|
||||
ChatCompletionAPIRequest? sent = null;
|
||||
var adapter = Adapter(mayAskForSequentialToolCalls, request => sent = request, ["[DONE]"]);
|
||||
await foreach (var _ in adapter.ExecuteRoundAsync(null, includeTools))
|
||||
{
|
||||
}
|
||||
|
||||
return JsonSerializer.Serialize(sent, ProviderJsonOptions.OPTIONS);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Runs the next round and returns the prompt of every usage it passed on.
|
||||
/// </summary>
|
||||
private static async Task<List<int>> Usages(ChatCompletionToolCallingAdapter<ChatCompletionAPIRequest> adapter)
|
||||
{
|
||||
var usages = new List<int>();
|
||||
await foreach (var streamEvent in adapter.ExecuteRoundAsync(null, true))
|
||||
if (streamEvent.Delta is { Usage.IsKnown: true } delta)
|
||||
usages.Add(delta.Usage.PromptTokens);
|
||||
|
||||
return usages;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Builds an adapter whose requests are answered by the given rounds, one after another.
|
||||
/// </summary>
|
||||
private static ChatCompletionToolCallingAdapter<ChatCompletionAPIRequest> Adapter(params string[][] rounds) => Adapter(true, _ => { }, rounds);
|
||||
|
||||
/// <summary>
|
||||
/// Builds an adapter whose requests are answered by the given rounds, and which hands every
|
||||
/// request it sends to the given observer.
|
||||
/// </summary>
|
||||
private static ChatCompletionToolCallingAdapter<ChatCompletionAPIRequest> Adapter(bool mayAskForSequentialToolCalls, Action<ChatCompletionAPIRequest> sent, params string[][] rounds)
|
||||
{
|
||||
var nextRound = 0;
|
||||
return new(
|
||||
(_, _, tools) => Task.FromResult(new ChatCompletionAPIRequest("model-a", [], true) { Tools = tools }),
|
||||
new TextMessage("You are a helpful assistant.", "system"),
|
||||
new Dictionary<string, object>(),
|
||||
[],
|
||||
mayAskForSequentialToolCalls,
|
||||
[],
|
||||
(request, token) =>
|
||||
{
|
||||
sent(request);
|
||||
return Lines(rounds[nextRound++], token);
|
||||
},
|
||||
_ => [],
|
||||
NullLogger.Instance);
|
||||
}
|
||||
|
||||
private static async IAsyncEnumerable<ServerSentEvent> Lines(string[] data, [EnumeratorCancellation] CancellationToken token = default)
|
||||
{
|
||||
foreach (var line in data)
|
||||
{
|
||||
token.ThrowIfCancellationRequested();
|
||||
yield return new ServerSentEvent($"data: {line}", line);
|
||||
}
|
||||
|
||||
await Task.CompletedTask;
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user