mirror of
https://github.com/MindWorkAI/AI-Studio.git
synced 2026-10-17 03:37:59 +00:00
Fixed tool calling with Groq and OpenRouter (#1003)
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 / Verify (push) Waiting to run
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 / 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 / Verify (push) Waiting to run
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
This commit is contained in:
1 parent
40ac215359
commit
0c585144e2
9 files changed
+516
-43
No files matched your search
@@ -1245,6 +1245,7 @@ public abstract class BaseProvider : IProvider, ISecretId
|
||||
/// <param name="requestPath">The request path, relative to the provider base URL.</param>
|
||||
/// <param name="headersAction">Optional additional headers to add.</param>
|
||||
/// <param name="mayAskForSequentialToolCalls">Whether a request which offers tools may ask for one call at a time. False for a provider which rejects the parallel_tool_calls parameter.</param>
|
||||
/// <param name="enforcesStrictToolSchemas">Whether the provider binds the model's tool calls to a strict schema. Only then are tools offered in strict mode; everywhere else a strict schema would only tell the model that every argument is required.</param>
|
||||
/// <param name="token">The cancellation token.</param>
|
||||
/// <typeparam name="TRequest">The request DTO type.</typeparam>
|
||||
/// <typeparam name="TDelta">The delta stream line type.</typeparam>
|
||||
@@ -1262,6 +1263,7 @@ public abstract class BaseProvider : IProvider, ISecretId
|
||||
string requestPath = "chat/completions",
|
||||
Action<HttpRequestHeaders>? headersAction = null,
|
||||
bool mayAskForSequentialToolCalls = true,
|
||||
bool enforcesStrictToolSchemas = false,
|
||||
[EnumeratorCancellation] CancellationToken token = default)
|
||||
where TRequest : ChatCompletionAPIRequest
|
||||
where TDelta : IResponseStreamLine
|
||||
@@ -1300,7 +1302,7 @@ public abstract class BaseProvider : IProvider, ISecretId
|
||||
if (runnableTools.Count > 0)
|
||||
{
|
||||
var adapter = new ChatCompletionToolCallingAdapter<TRequest>(requestFactory, systemPrompt, apiParameters,
|
||||
runnableTools.Select(x => ProviderToolAdapters.ToChatCompletionTool(x.Definition)).ToList(), mayAskForSequentialToolCalls, runnableTools,
|
||||
runnableTools.Select(x => ProviderToolAdapters.ToChatCompletionTool(x.Definition, enforcesStrictToolSchemas)).ToList(), mayAskForSequentialToolCalls, runnableTools,
|
||||
(requestDto, requestToken) => this.StreamChatCompletionRequest(requestDto, providerName, requestPath, requestedSecret, headersAction, requestToken),
|
||||
ChatCompletionSourceReader.Read<TDelta, TAnnotation>,
|
||||
this.logger);
|
||||
|
||||
@@ -168,6 +168,16 @@ public sealed class ProviderOpenAI() : BaseProvider(LLMProviders.OPEN_AI, new Ur
|
||||
},
|
||||
systemPromptRole: systemPromptRole,
|
||||
requestPath: "chat/completions",
|
||||
|
||||
//
|
||||
// OpenAI binds a strict function's arguments to its schema -- read on
|
||||
// 2026-09-24 at https://developers.openai.com/api/docs/guides/function-calling.
|
||||
// Most other hosts of this API do not: Groq applies strict mode to
|
||||
// response formats only (https://console.groq.com/docs/structured-outputs,
|
||||
// same day) and merely validates a tool call afterward, so it rejects
|
||||
// every call that leaves out an argument the strict schema requires:
|
||||
//
|
||||
enforcesStrictToolSchemas: true,
|
||||
token: token))
|
||||
yield return content;
|
||||
|
||||
|
||||
@@ -10,33 +10,47 @@ namespace AIStudio.Provider;
|
||||
/// <remarks>
|
||||
/// The definitions state a tool once, in plain JSON Schema. What differs per API is not only the
|
||||
/// field names but how an optional argument is expressed, which is why the OpenAI shapes convert
|
||||
/// the schema while Anthropic takes it as written.
|
||||
/// the schema while Anthropic takes it as written.<br/><br/>
|
||||
/// Strict mode is a promise of the host, not of the definition: only a host which binds the
|
||||
/// model's output to the schema keeps it. Everywhere else the model merely reads a schema that
|
||||
/// calls every argument required, and then either invents a value for an argument it meant to
|
||||
/// leave out, or the host rejects the call for lacking one. That is why the Chat Completions
|
||||
/// shape, which reaches every OpenAI-compatible host, only goes strict where the host says so.
|
||||
/// </remarks>
|
||||
public static class ProviderToolAdapters
|
||||
{
|
||||
/// <summary>
|
||||
/// Builds the nested function tool shape used by Chat Completions compatible APIs.
|
||||
/// </summary>
|
||||
public static object ToChatCompletionTool(ToolDefinition definition) => new
|
||||
/// <param name="definition">The tool to describe.</param>
|
||||
/// <param name="hostEnforcesStrict">Whether the host binds the model's tool calls to a strict schema. Only then is the tool sent in strict mode.</param>
|
||||
public static object ToChatCompletionTool(ToolDefinition definition, bool hostEnforcesStrict)
|
||||
{
|
||||
type = "function",
|
||||
function = new
|
||||
var isStrict = definition.Function.Strict && hostEnforcesStrict;
|
||||
return new
|
||||
{
|
||||
name = definition.Function.Name,
|
||||
description = definition.Function.DescriptionForLLM,
|
||||
parameters = ToOpenAIParameters(definition),
|
||||
strict = definition.Function.Strict,
|
||||
}
|
||||
};
|
||||
type = "function",
|
||||
function = new
|
||||
{
|
||||
name = definition.Function.Name,
|
||||
description = definition.Function.DescriptionForLLM,
|
||||
parameters = ToOpenAIParameters(definition, isStrict),
|
||||
strict = isStrict,
|
||||
}
|
||||
};
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Builds the flat function tool shape used by the OpenAI Responses API.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Only OpenAI speaks this API, and it enforces strict mode, so the definition alone decides.
|
||||
/// </remarks>
|
||||
public static ResponsesFunctionTool ToResponsesTool(ToolDefinition definition) => new()
|
||||
{
|
||||
Name = definition.Function.Name,
|
||||
Description = definition.Function.DescriptionForLLM,
|
||||
Parameters = ToOpenAIParameters(definition),
|
||||
Parameters = ToOpenAIParameters(definition, definition.Function.Strict),
|
||||
Strict = definition.Function.Strict,
|
||||
};
|
||||
|
||||
@@ -59,7 +73,7 @@ public static class ProviderToolAdapters
|
||||
/// <summary>
|
||||
/// The parameter schema for the OpenAI APIs, converted only when strict mode asks for it.
|
||||
/// </summary>
|
||||
private static System.Text.Json.JsonElement ToOpenAIParameters(ToolDefinition definition) => definition.Function.Strict
|
||||
private static System.Text.Json.JsonElement ToOpenAIParameters(ToolDefinition definition, bool isStrict) => isStrict
|
||||
? OpenAIStrictToolSchema.FromToolParameters(definition.Function.Parameters)
|
||||
: definition.Function.Parameters;
|
||||
}
|
||||
+99
-26
@@ -94,9 +94,28 @@ public sealed class WebSearchTool(IEnumerable<IWebSearchBackend> backends, WebPa
|
||||
private const string LIMIT_ARGUMENT = "limit";
|
||||
|
||||
private const string TIME_RANGE_DAY = "day";
|
||||
private const string TIME_RANGE_WEEK = "week";
|
||||
private const string TIME_RANGE_MONTH = "month";
|
||||
private const string TIME_RANGE_YEAR = "year";
|
||||
|
||||
/// <summary>
|
||||
/// The time ranges a search can be restricted to.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Those which both services with a time filter, SearXNG and Tavily, understand and take as
|
||||
/// they are. Tavily documents all four. SearXNG's API documentation names no week, but its code accepts
|
||||
/// one -- read on 2026-09-24 in parse_time_range of searx/webadapter.py. A model asked about
|
||||
/// "this week" wants exactly that, and without it, it asks for a week again and again.<br/><br/>
|
||||
/// The schema offers exactly these and the reader checks against them, so the two cannot drift
|
||||
/// apart.
|
||||
/// </remarks>
|
||||
private static readonly string[] TIME_RANGES = [TIME_RANGE_DAY, TIME_RANGE_WEEK, TIME_RANGE_MONTH, TIME_RANGE_YEAR];
|
||||
|
||||
/// <summary>
|
||||
/// How much of a wrongly passed argument an error message repeats back to the model.
|
||||
/// </summary>
|
||||
private const int MAX_ARGUMENT_ECHO_LENGTH = 40;
|
||||
|
||||
public string ImplementationKey => ToolSelectionRules.WEB_SEARCH_TOOL_ID;
|
||||
|
||||
/// <inheritdoc />
|
||||
@@ -118,7 +137,7 @@ public sealed class WebSearchTool(IEnumerable<IWebSearchBackend> backends, WebPa
|
||||
Parameters = ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString(QUERY_ARGUMENT, "The search query.")
|
||||
.OptionalString(LANGUAGE_ARGUMENT, "Optional IETF language tag restricting the search to one language, such as 'de-DE', 'en-US', or 'all' for no restriction. Leave it out to search in the language configured for this tool. Do not pass a language name such as 'German': search engines expect the tag and silently return nothing for anything else.")
|
||||
.OptionalEnum(TIME_RANGE_ARGUMENT, "Optional time range filter for the search.", TIME_RANGE_DAY, TIME_RANGE_MONTH, TIME_RANGE_YEAR)
|
||||
.OptionalEnum(TIME_RANGE_ARGUMENT, "Optional time range filter for the search.", TIME_RANGES)
|
||||
.OptionalInteger(PAGE_ARGUMENT, "Optional search result page number starting at 1.")
|
||||
.OptionalInteger(LIMIT_ARGUMENT, $"Optional maximum number of ranked result pages to retrieve and return. The hard maximum is {MAX_RESULTS}.")
|
||||
.Build(),
|
||||
@@ -482,14 +501,11 @@ public sealed class WebSearchTool(IEnumerable<IWebSearchBackend> backends, WebPa
|
||||
|
||||
public async Task<ToolExecutionResult> ExecuteAsync(JsonElement arguments, ToolExecutionContext context, CancellationToken token = default)
|
||||
{
|
||||
var query = ReadRequiredString(arguments, QUERY_ARGUMENT);
|
||||
var language = ReadOptionalString(arguments, LANGUAGE_ARGUMENT);
|
||||
var timeRange = ReadOptionalString(arguments, TIME_RANGE_ARGUMENT);
|
||||
var page = ReadOptionalPositiveInt(arguments, PAGE_ARGUMENT);
|
||||
var requestedLimit = ReadOptionalPositiveInt(arguments, LIMIT_ARGUMENT);
|
||||
|
||||
if (timeRange is not null && timeRange is not (TIME_RANGE_DAY or TIME_RANGE_MONTH or TIME_RANGE_YEAR))
|
||||
throw new ArgumentException($"Invalid time_range '{timeRange}'.");
|
||||
var query = ReadQuery(arguments);
|
||||
var language = ReadLanguage(arguments);
|
||||
var timeRange = ReadTimeRange(arguments);
|
||||
var page = ReadPage(arguments);
|
||||
var requestedLimit = ReadLimit(arguments);
|
||||
|
||||
language = string.IsNullOrWhiteSpace(language) ? context.SettingsValues.GetValueOrDefault(DEFAULT_LANGUAGE_SETTING) : language;
|
||||
var safeSearch = ReadSafeSearchPolicy(context.SettingsValues);
|
||||
@@ -807,42 +823,99 @@ public sealed class WebSearchTool(IEnumerable<IWebSearchBackend> backends, WebPa
|
||||
return result;
|
||||
}
|
||||
|
||||
private static string ReadRequiredString(JsonElement arguments, string propertyName)
|
||||
/// <summary>
|
||||
/// Reads the search query, the one argument the model always has to pass.
|
||||
/// </summary>
|
||||
internal static string ReadQuery(JsonElement arguments)
|
||||
{
|
||||
var value = ReadOptionalString(arguments, propertyName);
|
||||
if (string.IsNullOrWhiteSpace(value))
|
||||
throw new ArgumentException($"Missing required argument '{propertyName}'.");
|
||||
var query = ReadOptionalString(arguments, QUERY_ARGUMENT, whenLeftOut: null);
|
||||
if (string.IsNullOrWhiteSpace(query))
|
||||
throw new ArgumentException($"Missing required argument '{QUERY_ARGUMENT}'.");
|
||||
|
||||
return value;
|
||||
return query;
|
||||
}
|
||||
|
||||
private static string? ReadOptionalString(JsonElement arguments, string propertyName)
|
||||
/// <summary>
|
||||
/// Reads the language tag the model asked for, or null for the configured language.
|
||||
/// </summary>
|
||||
internal static string? ReadLanguage(JsonElement arguments) => ReadOptionalString(arguments, LANGUAGE_ARGUMENT, "to use the configured language");
|
||||
|
||||
/// <summary>
|
||||
/// Reads the time range the model asked for, or null for no restriction.
|
||||
/// </summary>
|
||||
internal static string? ReadTimeRange(JsonElement arguments)
|
||||
{
|
||||
if (!arguments.TryGetProperty(propertyName, out var value))
|
||||
if (!TryGetArgument(arguments, TIME_RANGE_ARGUMENT, out var value))
|
||||
return null;
|
||||
|
||||
return value.ValueKind switch
|
||||
{
|
||||
JsonValueKind.Null => null,
|
||||
JsonValueKind.String => value.GetString()?.Trim(),
|
||||
_ => throw new ArgumentException($"Argument '{propertyName}' must be a string."),
|
||||
};
|
||||
var timeRange = value.ValueKind is JsonValueKind.String ? value.GetString()?.Trim() : null;
|
||||
if (timeRange is null || !TIME_RANGES.Contains(timeRange, StringComparer.Ordinal))
|
||||
throw InvalidArgument(TIME_RANGE_ARGUMENT, value, $"one of {string.Join(", ", TIME_RANGES)}", "to search without a time restriction");
|
||||
|
||||
return timeRange;
|
||||
}
|
||||
|
||||
private static int? ReadOptionalPositiveInt(JsonElement arguments, string propertyName)
|
||||
/// <summary>
|
||||
/// Reads the result page the model asked for, or null for the first one.
|
||||
/// </summary>
|
||||
internal static int? ReadPage(JsonElement arguments) => ReadOptionalPositiveInt(arguments, PAGE_ARGUMENT, "to get the first page");
|
||||
|
||||
/// <summary>
|
||||
/// Reads how many results the model asked for, or null for the configured number.
|
||||
/// </summary>
|
||||
internal static int? ReadLimit(JsonElement arguments) => ReadOptionalPositiveInt(arguments, LIMIT_ARGUMENT, "to get as many results as configured");
|
||||
|
||||
/// <summary>
|
||||
/// Looks up an argument, treating null the same as leaving it out.
|
||||
/// </summary>
|
||||
private static bool TryGetArgument(JsonElement arguments, string propertyName, out JsonElement value) =>
|
||||
arguments.TryGetProperty(propertyName, out value) && value.ValueKind is not JsonValueKind.Null;
|
||||
|
||||
private static string? ReadOptionalString(JsonElement arguments, string propertyName, string? whenLeftOut)
|
||||
{
|
||||
if (!arguments.TryGetProperty(propertyName, out var value))
|
||||
if (!TryGetArgument(arguments, propertyName, out var value))
|
||||
return null;
|
||||
|
||||
if (value.ValueKind is JsonValueKind.Null)
|
||||
if (value.ValueKind is not JsonValueKind.String)
|
||||
throw InvalidArgument(propertyName, value, "a string", whenLeftOut);
|
||||
|
||||
return value.GetString()?.Trim();
|
||||
}
|
||||
|
||||
private static int? ReadOptionalPositiveInt(JsonElement arguments, string propertyName, string whenLeftOut)
|
||||
{
|
||||
if (!TryGetArgument(arguments, propertyName, out var value))
|
||||
return null;
|
||||
|
||||
if (value.ValueKind is not JsonValueKind.Number || !value.TryGetInt32(out var intValue) || intValue <= 0)
|
||||
throw new ArgumentException($"Argument '{propertyName}' must be a positive integer.");
|
||||
throw InvalidArgument(propertyName, value, "a positive integer", whenLeftOut);
|
||||
|
||||
return intValue;
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Builds the error a model gets for an argument it passed wrongly.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// The model reads this and tries again, so it says what arrived, what would have been right,
|
||||
/// and, for an optional argument, that leaving it out is always an option. A model which
|
||||
/// believes the argument has to be there otherwise keeps trying placeholders, and every attempt
|
||||
/// costs one of the tool calls an answer may make.
|
||||
/// </remarks>
|
||||
/// <param name="propertyName">The argument.</param>
|
||||
/// <param name="value">What the model passed, as it arrived.</param>
|
||||
/// <param name="expectation">What the argument must be, completing "must be ...".</param>
|
||||
/// <param name="whenLeftOut">What happens without the argument, completing "Leave it out ...", or null for a required one.</param>
|
||||
private static ArgumentException InvalidArgument(string propertyName, JsonElement value, string expectation, string? whenLeftOut)
|
||||
{
|
||||
var receivedValue = value.GetRawText();
|
||||
if (receivedValue.Length > MAX_ARGUMENT_ECHO_LENGTH)
|
||||
receivedValue = $"{receivedValue[..MAX_ARGUMENT_ECHO_LENGTH]}...";
|
||||
|
||||
var message = $"Argument '{propertyName}' must be {expectation}, but was {receivedValue}.";
|
||||
return new ArgumentException(whenLeftOut is null ? message : $"{message} Leave it out {whenLeftOut}.");
|
||||
}
|
||||
|
||||
private static string FormatQueryForLog(string query)
|
||||
{
|
||||
var singleLineQuery = query
|
||||
|
||||
@@ -8,6 +8,14 @@ public sealed class ToolFunctionDefinition
|
||||
|
||||
public string DescriptionForLLM { get; init; } = string.Empty;
|
||||
|
||||
/// <summary>
|
||||
/// Whether this tool may be offered in strict mode.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// The host has a say as well: a tool goes strict only where the host binds the model's calls
|
||||
/// to the schema, see ProviderToolAdapters. Setting this to false keeps a tool out of strict
|
||||
/// mode everywhere, for a schema which strict mode cannot express.
|
||||
/// </remarks>
|
||||
public bool Strict { get; init; } = true;
|
||||
|
||||
public JsonElement Parameters { get; init; }
|
||||
|
||||
@@ -0,0 +1,107 @@
|
||||
using System.Text.Json.Nodes;
|
||||
|
||||
using AIStudio.Provider.OpenAI;
|
||||
using AIStudio.Tools.ToolCallingSystem;
|
||||
|
||||
namespace AIStudio.Tests.Provider.ToolCalling;
|
||||
|
||||
/// <summary>
|
||||
/// Checks how a tool's parameter schema is translated into the form OpenAI's strict mode requires.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Strict mode wants every argument required and says "may be left out" by allowing null instead.
|
||||
/// The translation has to keep what the schema means: an argument the tool can do without stays
|
||||
/// one the model can leave empty, and an argument the tool needs stays exactly as it was.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class OpenAIStrictToolSchemaTests
|
||||
{
|
||||
[Test]
|
||||
public void AnOptionalArgumentMayBeNullInstead()
|
||||
{
|
||||
var properties = Converted(ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("query", "The search query.")
|
||||
.OptionalString("language", "The language.")
|
||||
.OptionalInteger("page", "The page."))["properties"]!;
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(Types(properties["language"]!), Is.EqualTo(new[] { "string", "null" }));
|
||||
Assert.That(Types(properties["page"]!), Is.EqualTo(new[] { "integer", "null" }));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AnOptionalChoiceOffersNullAmongItsValues()
|
||||
{
|
||||
var timeRange = Converted(ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("query", "The search query.")
|
||||
.OptionalEnum("time_range", "The time range.", "day", "month", "year"))["properties"]!["time_range"]!;
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(Types(timeRange), Is.EqualTo(new[] { "string", "null" }));
|
||||
Assert.That(timeRange["enum"]!.AsArray().Select(value => value?.GetValue<string>()), Is.EqualTo(new[] { null, "day", "month", "year" }));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ARequiredArgumentStaysAsItIs()
|
||||
{
|
||||
var query = Converted(ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("query", "The search query.")
|
||||
.OptionalString("language", "The language."))["properties"]!["query"]!;
|
||||
|
||||
Assert.That(query["type"]!.GetValue<string>(), Is.EqualTo("string"));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void EveryArgumentIsRequiredInTheOrderOfTheProperties()
|
||||
{
|
||||
//
|
||||
// The order has to be stable across requests, because prompt caching depends on it. The
|
||||
// optional argument comes first here, so an order taken from the old required list would
|
||||
// show.
|
||||
//
|
||||
var schema = Converted(ToolParameterSchemaBuilder.Create()
|
||||
.OptionalString("language", "The language.")
|
||||
.RequiredString("query", "The search query."));
|
||||
|
||||
Assert.That(schema["required"]!.AsArray().Select(name => name!.GetValue<string>()), Is.EqualTo(new[] { "language", "query" }));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ExtraArgumentsStayRefused()
|
||||
{
|
||||
var schema = Converted(ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("query", "The search query.")
|
||||
.OptionalString("language", "The language."));
|
||||
|
||||
Assert.That(schema["additionalProperties"]!.GetValue<bool>(), Is.False);
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ASchemaWithoutOptionalArgumentsIsLeftUntouched()
|
||||
{
|
||||
var parameters = ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("url", "The address of the page.")
|
||||
.Build();
|
||||
|
||||
var converted = OpenAIStrictToolSchema.FromToolParameters(parameters);
|
||||
Assert.That(JsonNode.DeepEquals(JsonNode.Parse(converted.GetRawText()), JsonNode.Parse(parameters.GetRawText())), Is.True);
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// Builds the schema and returns it the way strict mode receives it.
|
||||
/// </summary>
|
||||
private static JsonNode Converted(ToolParameterSchemaBuilder builder) => JsonNode.Parse(OpenAIStrictToolSchema.FromToolParameters(builder.Build()).GetRawText())!;
|
||||
|
||||
/// <summary>
|
||||
/// Reads the types a property allows, whether it names one or several.
|
||||
/// </summary>
|
||||
private static string[] Types(JsonNode property) => property["type"] switch
|
||||
{
|
||||
JsonArray types => types.Select(type => type!.GetValue<string>()).ToArray(),
|
||||
var type => [type!.GetValue<string>()],
|
||||
};
|
||||
}
|
||||
@@ -0,0 +1,85 @@
|
||||
using System.Text.Json;
|
||||
using System.Text.Json.Nodes;
|
||||
|
||||
using AIStudio.Provider;
|
||||
using AIStudio.Tools.ToolCallingSystem;
|
||||
|
||||
namespace AIStudio.Tests.Provider.ToolCalling;
|
||||
|
||||
/// <summary>
|
||||
/// Checks when a tool reaches a Chat Completions host in strict mode, and when as written.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// Strict mode only holds where the host binds the model's calls to the schema. Anywhere else, the
|
||||
/// model just reads a schema which calls every argument required: Groq then rejects each call that
|
||||
/// leaves one out, and other models fill the gap with a placeholder the tool has to refuse. Such a
|
||||
/// host therefore gets the schema the way the tool wrote it.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class ProviderToolAdaptersTests
|
||||
{
|
||||
[Test]
|
||||
public void AHostWhichDoesNotEnforceStrictModeGetsTheSchemaAsWritten()
|
||||
{
|
||||
var definition = WebSearchLike();
|
||||
var function = Sent(definition, hostEnforcesStrict: false);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(function["strict"]!.GetValue<bool>(), Is.False);
|
||||
Assert.That(JsonNode.DeepEquals(function["parameters"], JsonNode.Parse(definition.Function.Parameters.GetRawText())), Is.True);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AHostWhichEnforcesStrictModeGetsTheConvertedSchema()
|
||||
{
|
||||
var function = Sent(WebSearchLike(), hostEnforcesStrict: true);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(function["strict"]!.GetValue<bool>(), Is.True);
|
||||
Assert.That(function["parameters"]!["required"]!.AsArray().Select(name => name!.GetValue<string>()), Is.EqualTo(new[] { "query", "page" }));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AToolWhichOptsOutStaysOutOfStrictModeEverywhere()
|
||||
{
|
||||
var definition = WebSearchLike(isStrict: false);
|
||||
var function = Sent(definition, hostEnforcesStrict: true);
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(function["strict"]!.GetValue<bool>(), Is.False);
|
||||
Assert.That(JsonNode.DeepEquals(function["parameters"], JsonNode.Parse(definition.Function.Parameters.GetRawText())), Is.True);
|
||||
});
|
||||
}
|
||||
|
||||
/// <summary>
|
||||
/// A tool with one required and one optional argument, the shape the web search has.
|
||||
/// </summary>
|
||||
private static ToolDefinition WebSearchLike(bool isStrict = true) => new()
|
||||
{
|
||||
Id = "web_search",
|
||||
Function = new()
|
||||
{
|
||||
Name = "web_search",
|
||||
DescriptionForLLM = "Search the web.",
|
||||
Strict = isStrict,
|
||||
Parameters = ToolParameterSchemaBuilder.Create()
|
||||
.RequiredString("query", "The search query.")
|
||||
.OptionalInteger("page", "The page, starting at 1.")
|
||||
.Build(),
|
||||
},
|
||||
};
|
||||
|
||||
/// <summary>
|
||||
/// Returns the function object of the tool as it goes over the wire.
|
||||
/// </summary>
|
||||
private static JsonNode Sent(ToolDefinition definition, bool hostEnforcesStrict)
|
||||
{
|
||||
var tool = ProviderToolAdapters.ToChatCompletionTool(definition, hostEnforcesStrict);
|
||||
return JsonNode.Parse(JsonSerializer.Serialize(tool, ProviderJsonOptions.OPTIONS))!["function"]!;
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,170 @@
|
||||
using System.Text.Json;
|
||||
|
||||
using AIStudio.Tools.ToolCallingSystem.ToolCallingImplementations.WebSearch;
|
||||
|
||||
namespace AIStudio.Tests.Tools.ToolCalling;
|
||||
|
||||
/// <summary>
|
||||
/// Checks how the web search reads the arguments a model passes, and what it says about wrong ones.
|
||||
/// </summary>
|
||||
/// <remarks>
|
||||
/// A wrong argument is refused rather than guessed at: a placeholder such as 0 is not a page, and
|
||||
/// quietly reading it as "no page" would do something the model did not ask for. What makes the
|
||||
/// refusal work is its message. The model reads it and tries again, so it has to say what arrived,
|
||||
/// what would have been right, and that an optional argument may simply be left out. A model
|
||||
/// which believes it has to pass one otherwise keeps trying placeholders, and every attempt costs
|
||||
/// one of the tool calls an answer may make.
|
||||
/// </remarks>
|
||||
[TestFixture]
|
||||
public sealed class WebSearchToolArgumentTests
|
||||
{
|
||||
[Test]
|
||||
public void AnArgumentLeftOutIsNotSet()
|
||||
{
|
||||
var arguments = Arguments("""{"query":"weather"}""");
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(WebSearchTool.ReadLanguage(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadTimeRange(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadPage(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadLimit(arguments), Is.Null);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ANullArgumentIsTheSameAsOneLeftOut()
|
||||
{
|
||||
var arguments = Arguments("""{"query":"weather","language":null,"time_range":null,"page":null,"limit":null}""");
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(WebSearchTool.ReadLanguage(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadTimeRange(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadPage(arguments), Is.Null);
|
||||
Assert.That(WebSearchTool.ReadLimit(arguments), Is.Null);
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ValidArgumentsComeThrough()
|
||||
{
|
||||
var arguments = Arguments("""{"query":" weather ","language":"de-DE","time_range":"day","page":2,"limit":7}""");
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(WebSearchTool.ReadQuery(arguments), Is.EqualTo("weather"));
|
||||
Assert.That(WebSearchTool.ReadLanguage(arguments), Is.EqualTo("de-DE"));
|
||||
Assert.That(WebSearchTool.ReadTimeRange(arguments), Is.EqualTo("day"));
|
||||
Assert.That(WebSearchTool.ReadPage(arguments), Is.EqualTo(2));
|
||||
Assert.That(WebSearchTool.ReadLimit(arguments), Is.EqualTo(7));
|
||||
});
|
||||
}
|
||||
|
||||
[TestCase("0")]
|
||||
[TestCase("-1")]
|
||||
[TestCase("2.5")]
|
||||
[TestCase("\"5\"")]
|
||||
[TestCase("true")]
|
||||
public void AWrongPageIsRefusedWithWhatArrived(string value)
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadPage(Arguments($$"""{"query":"weather","page":{{value}}}""")));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(message, Does.Contain("'page'").And.Contain("a positive integer"));
|
||||
Assert.That(message, Does.Contain($"but was {value}."), "Without the value, the model cannot tell which of its arguments the tool means.");
|
||||
Assert.That(message, Does.Contain("Leave it out"), "The way out a model needs when it believes the argument is required.");
|
||||
});
|
||||
}
|
||||
|
||||
[TestCase("0")]
|
||||
[TestCase("-1")]
|
||||
[TestCase("\"5\"")]
|
||||
public void AWrongLimitIsRefusedWithWhatArrived(string value)
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadLimit(Arguments($$"""{"query":"weather","limit":{{value}}}""")));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(message, Does.Contain("'limit'").And.Contain("a positive integer"));
|
||||
Assert.That(message, Does.Contain($"but was {value}."));
|
||||
Assert.That(message, Does.Contain("Leave it out"));
|
||||
});
|
||||
}
|
||||
|
||||
[TestCase("day")]
|
||||
[TestCase("week")]
|
||||
[TestCase("month")]
|
||||
[TestCase("year")]
|
||||
public void EveryOfferedTimeRangeIsAccepted(string timeRange)
|
||||
{
|
||||
//
|
||||
// The week is the one a model asks for when the user says "this week". Both services with
|
||||
// a time filter understand it, and refusing it only made the model try it again and again.
|
||||
//
|
||||
Assert.That(WebSearchTool.ReadTimeRange(Arguments($$"""{"query":"weather","time_range":"{{timeRange}}"}""")), Is.EqualTo(timeRange));
|
||||
}
|
||||
|
||||
[TestCase("\"\"")]
|
||||
[TestCase("\"Day\"")]
|
||||
[TestCase("\"decade\"")]
|
||||
[TestCase("5")]
|
||||
public void AWrongTimeRangeIsRefusedWithTheValuesThatWouldDo(string value)
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadTimeRange(Arguments($$"""{"query":"weather","time_range":{{value}}}""")));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(message, Does.Contain("'time_range'").And.Contain("one of day, week, month, year"));
|
||||
Assert.That(message, Does.Contain($"but was {value}."));
|
||||
Assert.That(message, Does.Contain("Leave it out"));
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ALanguageWhichIsNoStringIsRefused()
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadLanguage(Arguments("""{"query":"weather","language":5}""")));
|
||||
Assert.That(message, Does.Contain("'language'").And.Contain("but was 5.").And.Contain("Leave it out"));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AMissingQueryIsRefused()
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadQuery(Arguments("""{"page":1}""")));
|
||||
Assert.That(message, Does.Contain("Missing required argument 'query'"));
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void AQueryWhichIsNoStringIsRefusedWithoutOfferingToLeaveItOut()
|
||||
{
|
||||
var message = Refusal(() => WebSearchTool.ReadQuery(Arguments("""{"query":42}""")));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(message, Does.Contain("'query'").And.Contain("but was 42."));
|
||||
Assert.That(message, Does.Not.Contain("Leave it out"), "The query is required, so leaving it out is no way out.");
|
||||
});
|
||||
}
|
||||
|
||||
[Test]
|
||||
public void ALongValueIsShortenedInTheMessage()
|
||||
{
|
||||
var longValue = new string('x', 500);
|
||||
var message = Refusal(() => WebSearchTool.ReadTimeRange(Arguments($$"""{"query":"weather","time_range":"{{longValue}}"}""")));
|
||||
|
||||
Assert.Multiple(() =>
|
||||
{
|
||||
Assert.That(message, Does.Not.Contain(longValue), "The model sent the value itself; repeating all of it back only costs tokens.");
|
||||
Assert.That(message, Does.Contain("..."));
|
||||
});
|
||||
}
|
||||
|
||||
private static JsonElement Arguments(string json) => JsonSerializer.Deserialize<JsonElement>(json);
|
||||
|
||||
/// <summary>
|
||||
/// Runs a reader which has to refuse its argument and returns what it said.
|
||||
/// </summary>
|
||||
private static string Refusal(TestDelegate read) => Assert.Throws<ArgumentException>(read)!.Message;
|
||||
}
|
||||
Reference in new issue
Block a user