@agentica/benchmark 0.8.3 → 0.9.0-dev.20250302

package/src/structures/IAgenticaSelectBenchmarkEvent.ts

@@ -1,110 +1,114 @@
-import {
-  IAgenticaOperationSelection,
-  IAgenticaPrompt,
-  IAgenticaTokenUsage,
-} from "@agentica/core";
-
-import { IAgenticaSelectBenchmarkScenario } from "./IAgenticaSelectBenchmarkScenario";
-
-/**
- * Event of LLM function selection benchmark.
- *
- * `IAgenticaSelectBenchmarkEvent` is an union type of the events occurred
- * during the LLM function selection benchmark, representing one phase of
- * the benchmark testing about a scenario.
- *
- * In other words, when {@link AgenticaSelectBenchmark} executes the
- * benchmark, it will run the benchmark will test a scenario repeately with
- * the given configuration {@link AgenticaSelectBenchmark.IConfig.repeat}.
- * And in the repeated benchmark about a scenario,
- * `IAgenticaSelectBenchmarkEvent` is one of the repeated testing.
- *
- * For reference, `IAgenticaSelectBenchmarkEvent` is categorized into three
- * types: `success`, `failure`, and `error`. The `success` means the
- * benchmark testing is fully meet the expected scenario, and `failure`
- * means that the `selector` had not selected the expected operations. The
- * last type, `error`, means that an error had been occurred during the
- * benchmark testing.
- *
- * @author Samchon
- */
-export type IAgenticaSelectBenchmarkEvent =
-  | IAgenticaSelectBenchmarkEvent.ISuccess
-  | IAgenticaSelectBenchmarkEvent.IFailure
-  | IAgenticaSelectBenchmarkEvent.IError;
-export namespace IAgenticaSelectBenchmarkEvent {
-  /**
-   * Success event type.
-   *
-   * The `success` event type represents that the benchmark testing is
-   * fully meet the expected scenario.
-   */
-  export interface ISuccess extends IEventBase<"success"> {
-    /**
-     * Usage of the token during the benchmark.
-     */
-    usage: IAgenticaTokenUsage;
-
-    /**
-     * Selected operations in the benchmark.
-     */
-    selected: IAgenticaOperationSelection[];
-
-    /**
-     * Prompt messages from the assistant.
-     */
-    assistantPrompts: IAgenticaPrompt.IText<"assistant">[];
-  }
-
-  /**
-   * Failure event type.
-   *
-   * The `failure` event type represents that the `selector` had not
-   * selected the expected scenario in the benchmark testing.
-   */
-  export interface IFailure extends IEventBase<"failure"> {
-    /**
-     * Usage of the token during the benchmark.
-     */
-    usage: IAgenticaTokenUsage;
-
-    /**
-     * Selected operations in the benchmark.
-     */
-    selected: IAgenticaOperationSelection[];
-
-    /**
-     * Prompt messages from the assistant.
-     */
-    assistantPrompts: IAgenticaPrompt.IText<"assistant">[];
-  }
-
-  export interface IError extends IEventBase<"error"> {
-    /**
-     * Error occurred during the benchmark.
-     */
-    error: unknown;
-  }
-
-  interface IEventBase<Type extends string> {
-    /**
-     * Discriminant type.
-     */
-    type: Type;
-
-    /**
-     * Expected scenario.
-     */
-    scenario: IAgenticaSelectBenchmarkScenario;
-
-    /**
-     * When the benchmark testing started.
-     */
-    started_at: Date;
-
-    /**
-     * When the benchmark testing completed.
-     */
-    completed_at: Date;
-  }
-}
+import {
+  IAgenticaOperationSelection,
+  IAgenticaPrompt,
+  IAgenticaTokenUsage,
+} from "@agentica/core";
+import { ILlmSchema } from "@samchon/openapi";
+
+import { IAgenticaSelectBenchmarkScenario } from "./IAgenticaSelectBenchmarkScenario";
+
+/**
+ * Event of LLM function selection benchmark.
+ *
+ * `IAgenticaSelectBenchmarkEvent` is an union type of the events occurred
+ * during the LLM function selection benchmark, representing one phase of
+ * the benchmark testing about a scenario.
+ *
+ * In other words, when {@link AgenticaSelectBenchmark} executes the
+ * benchmark, it will run the benchmark will test a scenario repeately with
+ * the given configuration {@link AgenticaSelectBenchmark.IConfig.repeat}.
+ * And in the repeated benchmark about a scenario,
+ * `IAgenticaSelectBenchmarkEvent` is one of the repeated testing.
+ *
+ * For reference, `IAgenticaSelectBenchmarkEvent` is categorized into three
+ * types: `success`, `failure`, and `error`. The `success` means the
+ * benchmark testing is fully meet the expected scenario, and `failure`
+ * means that the `selector` had not selected the expected operations. The
+ * last type, `error`, means that an error had been occurred during the
+ * benchmark testing.
+ *
+ * @author Samchon
+ */
+export type IAgenticaSelectBenchmarkEvent<Model extends ILlmSchema.Model> =
+  | IAgenticaSelectBenchmarkEvent.ISuccess<Model>
+  | IAgenticaSelectBenchmarkEvent.IFailure<Model>
+  | IAgenticaSelectBenchmarkEvent.IError<Model>;
+export namespace IAgenticaSelectBenchmarkEvent {
+  /**
+   * Success event type.
+   *
+   * The `success` event type represents that the benchmark testing is
+   * fully meet the expected scenario.
+   */
+  export interface ISuccess<Model extends ILlmSchema.Model>
+    extends IEventBase<"success", Model> {
+    /**
+     * Usage of the token during the benchmark.
+     */
+    usage: IAgenticaTokenUsage;
+
+    /**
+     * Selected operations in the benchmark.
+     */
+    selected: IAgenticaOperationSelection<Model>[];
+
+    /**
+     * Prompt messages from the assistant.
+     */
+    assistantPrompts: IAgenticaPrompt.IText<"assistant">[];
+  }
+
+  /**
+   * Failure event type.
+   *
+   * The `failure` event type represents that the `selector` had not
+   * selected the expected scenario in the benchmark testing.
+   */
+  export interface IFailure<Model extends ILlmSchema.Model>
+    extends IEventBase<"failure", Model> {
+    /**
+     * Usage of the token during the benchmark.
+     */
+    usage: IAgenticaTokenUsage;
+
+    /**
+     * Selected operations in the benchmark.
+     */
+    selected: IAgenticaOperationSelection<Model>[];
+
+    /**
+     * Prompt messages from the assistant.
+     */
+    assistantPrompts: IAgenticaPrompt.IText<"assistant">[];
+  }
+
+  export interface IError<Model extends ILlmSchema.Model>
+    extends IEventBase<"error", Model> {
+    /**
+     * Error occurred during the benchmark.
+     */
+    error: unknown;
+  }
+
+  interface IEventBase<Type extends string, Model extends ILlmSchema.Model> {
+    /**
+     * Discriminant type.
+     */
+    type: Type;
+
+    /**
+     * Expected scenario.
+     */
+    scenario: IAgenticaSelectBenchmarkScenario<Model>;
+
+    /**
+     * When the benchmark testing started.
+     */
+    started_at: Date;
+
+    /**
+     * When the benchmark testing completed.
+     */
+    completed_at: Date;
+  }
+}

package/src/structures/IAgenticaSelectBenchmarkResult.ts

@@ -1,69 +1,72 @@
-import { IAgenticaTokenUsage } from "@agentica/core";
-
-import { IAgenticaSelectBenchmarkEvent } from "./IAgenticaSelectBenchmarkEvent";
-import { IAgenticaSelectBenchmarkScenario } from "./IAgenticaSelectBenchmarkScenario";
-
-/**
- * Result of the LLM function selection benchmark.
- *
- * `IAgenticaSelectBenchmarkResult` is a structure representing the result
- * of the LLM function selection benchmark executed by the
- * {@link AgenticaSelectBenchmark.execute execute} function.
- *
- * It contains every experiment results for each scenario, and aggregated
- * LLM token cost in the benchmark process.
- *
- * In each scenario, as the benchmark program experiments multiple times
- * about a scenario, it will contain multiple events. Also, because of the
- * characteristics of the LLM which is not predictable, the result can be
- * different in each event.
- *
- * @author Samchon
- */
-export interface IAgenticaSelectBenchmarkResult {
-  /**
-   * Experiments for each scenario.
-   */
-  experiments: IAgenticaSelectBenchmarkResult.IExperiment[];
-
-  /**
-   * Aggregated token usage information.
-   */
-  usage: IAgenticaTokenUsage;
-
-  /**
-   * Start time of the benchmark.
-   */
-  started_at: Date;
-
-  /**
-   * End time of the benchmark.
-   */
-  completed_at: Date;
-}
-export namespace IAgenticaSelectBenchmarkResult {
-  /**
-   * Experiment result about a scenario.
-   */
-  export interface IExperiment {
-    /**
-     * Expected scenario.
-     */
-    scenario: IAgenticaSelectBenchmarkScenario;
-
-    /**
-     * Events occurred during the benchmark in the scenario.
-     *
-     * When benchmarking a scenario, {@link AgenticaSelectBenchmark} will
-     * test a scenario multiple times with the given
-     * {@link AgenticaSelectBenchmark.IConfig.repeat repeat} count.
-     * And the event is one of the repeated benchmark results.
-     */
-    events: IAgenticaSelectBenchmarkEvent[];
-
-    /**
-     * LLM token usage information.
-     */
-    usage: IAgenticaTokenUsage;
-  }
-}
+import { IAgenticaTokenUsage } from "@agentica/core";
+import { ILlmSchema } from "@samchon/openapi";
+
+import { IAgenticaSelectBenchmarkEvent } from "./IAgenticaSelectBenchmarkEvent";
+import { IAgenticaSelectBenchmarkScenario } from "./IAgenticaSelectBenchmarkScenario";
+
+/**
+ * Result of the LLM function selection benchmark.
+ *
+ * `IAgenticaSelectBenchmarkResult` is a structure representing the result
+ * of the LLM function selection benchmark executed by the
+ * {@link AgenticaSelectBenchmark.execute execute} function.
+ *
+ * It contains every experiment results for each scenario, and aggregated
+ * LLM token cost in the benchmark process.
+ *
+ * In each scenario, as the benchmark program experiments multiple times
+ * about a scenario, it will contain multiple events. Also, because of the
+ * characteristics of the LLM which is not predictable, the result can be
+ * different in each event.
+ *
+ * @author Samchon
+ */
+export interface IAgenticaSelectBenchmarkResult<
+  Model extends ILlmSchema.Model,
+> {
+  /**
+   * Experiments for each scenario.
+   */
+  experiments: IAgenticaSelectBenchmarkResult.IExperiment<Model>[];
+
+  /**
+   * Aggregated token usage information.
+   */
+  usage: IAgenticaTokenUsage;
+
+  /**
+   * Start time of the benchmark.
+   */
+  started_at: Date;
+
+  /**
+   * End time of the benchmark.
+   */
+  completed_at: Date;
+}
+export namespace IAgenticaSelectBenchmarkResult {
+  /**
+   * Experiment result about a scenario.
+   */
+  export interface IExperiment<Model extends ILlmSchema.Model> {
+    /**
+     * Expected scenario.
+     */
+    scenario: IAgenticaSelectBenchmarkScenario<Model>;
+
+    /**
+     * Events occurred during the benchmark in the scenario.
+     *
+     * When benchmarking a scenario, {@link AgenticaSelectBenchmark} will
+     * test a scenario multiple times with the given
+     * {@link AgenticaSelectBenchmark.IConfig.repeat repeat} count.
+     * And the event is one of the repeated benchmark results.
+     */
+    events: IAgenticaSelectBenchmarkEvent<Model>[];
+
+    /**
+     * LLM token usage information.
+     */
+    usage: IAgenticaTokenUsage;
+  }
+}

package/src/structures/IAgenticaSelectBenchmarkScenario.ts

@@ -1,39 +1,43 @@
-import { IAgenticaBenchmarkExpected } from "./IAgenticaBenchmarkExpected";
-
-/**
- * Scenario of function selection.
- *
- * `IAgenticaSelectBenchmarkScenario` is a data structure which
- * represents a function selection benchmark scenario. It contains two
- * properties; {@linkk text} and {@link operations}.
- *
- * The {@link text} means the conversation text from the user, and
- * the other {@link operations} are the expected operations that
- * should be selected by the `selector` agent through the {@link text}
- * conversation.
- *
- * @author Samchon
- */
-export interface IAgenticaSelectBenchmarkScenario {
-  /**
-   * Name of the scenario.
-   *
-   * It must be unique within the benchmark scenarios.
-   */
-  name: string;
-
-  /**
-   * The prompt text from user.
-   */
-  text: string;
-
-  /**
-   * Expected function selection sequence.
-   *
-   * Sequence of operations (API operation or class function) that
-   * should be selected by the `selector` agent from the user's
-   * {@link text} conversation for the LLM (Large Language Model)
-   * function selection.
-   */
-  expected: IAgenticaBenchmarkExpected;
-}
+import { ILlmSchema } from "@samchon/openapi";
+
+import { IAgenticaBenchmarkExpected } from "./IAgenticaBenchmarkExpected";
+
+/**
+ * Scenario of function selection.
+ *
+ * `IAgenticaSelectBenchmarkScenario` is a data structure which
+ * represents a function selection benchmark scenario. It contains two
+ * properties; {@linkk text} and {@link operations}.
+ *
+ * The {@link text} means the conversation text from the user, and
+ * the other {@link operations} are the expected operations that
+ * should be selected by the `selector` agent through the {@link text}
+ * conversation.
+ *
+ * @author Samchon
+ */
+export interface IAgenticaSelectBenchmarkScenario<
+  Model extends ILlmSchema.Model,
+> {
+  /**
+   * Name of the scenario.
+   *
+   * It must be unique within the benchmark scenarios.
+   */
+  name: string;
+
+  /**
+   * The prompt text from user.
+   */
+  text: string;
+
+  /**
+   * Expected function selection sequence.
+   *
+   * Sequence of operations (API operation or class function) that
+   * should be selected by the `selector` agent from the user's
+   * {@link text} conversation for the LLM (Large Language Model)
+   * function selection.
+   */
+  expected: IAgenticaBenchmarkExpected<Model>;
+}

package/src/utils/MathUtil.ts

@@ -1,3 +1,3 @@
-export namespace MathUtil {
-  export const round = (value: number): number => Math.floor(value * 100) / 100;
-}
+export namespace MathUtil {
+  export const round = (value: number): number => Math.floor(value * 100) / 100;
+}