modelfusion 0.30.1 → 0.32.0

package/model-provider/openai/OpenAITextGenerationModel.cjs

@@ -21,55 +21,72 @@ const TikTokenTokenizer_js_1 = require("./TikTokenTokenizer.cjs");
  * @see https://openai.com/pricing
  */
 exports.OPENAI_TEXT_GENERATION_MODELS = {
+    "gpt-3.5-turbo-instruct": {
+        contextWindowSize: 4097,
+        promptTokenCostInMillicents: 0.15,
+        completionTokenCostInMillicents: 0.2,
+    },
     "davinci-002": {
         contextWindowSize: 16384,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
         fineTunedTokenCostInMillicents: 1.2,
     },
     "babbage-002": {
         contextWindowSize: 16384,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
         fineTunedTokenCostInMillicents: 0.16,
     },
     "text-davinci-003": {
         contextWindowSize: 4096,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "text-davinci-002": {
         contextWindowSize: 4096,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "code-davinci-002": {
         contextWindowSize: 8000,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     davinci: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "text-curie-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
     },
     curie: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
     },
     "text-babbage-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.05,
+        promptTokenCostInMillicents: 0.05,
+        completionTokenCostInMillicents: 0.05,
     },
     babbage: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.05,
+        promptTokenCostInMillicents: 0.05,
+        completionTokenCostInMillicents: 0.05,
     },
     "text-ada-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
     },
     ada: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
     },
 };
 function getOpenAITextGenerationModelInformation(model) {
@@ -80,7 +97,8 @@ function getOpenAITextGenerationModelInformation(model) {
             baseModel: model,
             isFineTuned: false,
             contextWindowSize: baseModelInformation.contextWindowSize,
-            tokenCostInMillicents: baseModelInformation.tokenCostInMillicents,
+            promptTokenCostInMillicents: baseModelInformation.promptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.completionTokenCostInMillicents,
         };
     }
     // Extract the base model from the fine-tuned model:
@@ -92,7 +110,8 @@ function getOpenAITextGenerationModelInformation(model) {
             baseModel: baseModel,
             isFineTuned: true,
             contextWindowSize: baseModelInformation.contextWindowSize,
-            tokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+            promptTokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
         };
     }
     throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
@@ -102,8 +121,13 @@ const isOpenAITextGenerationModel = (model) => model in exports.OPENAI_TEXT_GENE
     model.startsWith("ft:davinci-002:") ||
     model.startsWith("ft:babbage-002:");
 exports.isOpenAITextGenerationModel = isOpenAITextGenerationModel;
-const calculateOpenAITextGenerationCostInMillicents = ({ model, response, }) => response.usage.total_tokens *
-    getOpenAITextGenerationModelInformation(model).tokenCostInMillicents;
+const calculateOpenAITextGenerationCostInMillicents = ({ model, response, }) => {
+    const modelInformation = getOpenAITextGenerationModelInformation(model);
+    return (response.usage.prompt_tokens *
+        modelInformation.promptTokenCostInMillicents +
+        response.usage.completion_tokens *
+            modelInformation.completionTokenCostInMillicents);
+};
 exports.calculateOpenAITextGenerationCostInMillicents = calculateOpenAITextGenerationCostInMillicents;
 /**
  * Create a text generation model that calls the OpenAI text completion API.
@@ -112,7 +136,7 @@ exports.calculateOpenAITextGenerationCostInMillicents = calculateOpenAITextGener
  *
  * @example
  * const model = new OpenAITextGenerationModel({
- *   model: "text-davinci-003",
+ *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
  *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),

package/model-provider/openai/OpenAITextGenerationModel.d.ts

@@ -14,62 +14,80 @@ import { TikTokenTokenizer } from "./TikTokenTokenizer.js";
  * @see https://openai.com/pricing
  */
 export declare const OPENAI_TEXT_GENERATION_MODELS: {
+    "gpt-3.5-turbo-instruct": {
+        contextWindowSize: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
+    };
     "davinci-002": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
         fineTunedTokenCostInMillicents: number;
     };
     "babbage-002": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
         fineTunedTokenCostInMillicents: number;
     };
     "text-davinci-003": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     "text-davinci-002": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     "code-davinci-002": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     davinci: {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     "text-curie-001": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     curie: {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     "text-babbage-001": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     babbage: {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     "text-ada-001": {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
     ada: {
         contextWindowSize: number;
-        tokenCostInMillicents: number;
+        promptTokenCostInMillicents: number;
+        completionTokenCostInMillicents: number;
     };
 };
 export declare function getOpenAITextGenerationModelInformation(model: OpenAITextGenerationModelType): {
     baseModel: OpenAITextGenerationBaseModelType;
     isFineTuned: boolean;
     contextWindowSize: number;
-    tokenCostInMillicents: number;
+    promptTokenCostInMillicents: number;
+    completionTokenCostInMillicents: number;
 };
 type FineTuneableOpenAITextGenerationModelType = "davinci-002" | "babbage-002";
 type FineTunedOpenAITextGenerationModelType = `ft:${FineTuneableOpenAITextGenerationModelType}:${string}:${string}:${string}`;
@@ -106,7 +124,7 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
  *
  * @example
  * const model = new OpenAITextGenerationModel({
- *   model: "text-davinci-003",
+ *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
  *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),

package/model-provider/openai/OpenAITextGenerationModel.js

@@ -15,55 +15,72 @@ import { TikTokenTokenizer } from "./TikTokenTokenizer.js";
  * @see https://openai.com/pricing
  */
 export const OPENAI_TEXT_GENERATION_MODELS = {
+    "gpt-3.5-turbo-instruct": {
+        contextWindowSize: 4097,
+        promptTokenCostInMillicents: 0.15,
+        completionTokenCostInMillicents: 0.2,
+    },
     "davinci-002": {
         contextWindowSize: 16384,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
         fineTunedTokenCostInMillicents: 1.2,
     },
     "babbage-002": {
         contextWindowSize: 16384,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
         fineTunedTokenCostInMillicents: 0.16,
     },
     "text-davinci-003": {
         contextWindowSize: 4096,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "text-davinci-002": {
         contextWindowSize: 4096,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "code-davinci-002": {
         contextWindowSize: 8000,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     davinci: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 2,
+        promptTokenCostInMillicents: 2,
+        completionTokenCostInMillicents: 2,
     },
     "text-curie-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
     },
     curie: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.2,
+        promptTokenCostInMillicents: 0.2,
+        completionTokenCostInMillicents: 0.2,
     },
     "text-babbage-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.05,
+        promptTokenCostInMillicents: 0.05,
+        completionTokenCostInMillicents: 0.05,
     },
     babbage: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.05,
+        promptTokenCostInMillicents: 0.05,
+        completionTokenCostInMillicents: 0.05,
     },
     "text-ada-001": {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
     },
     ada: {
         contextWindowSize: 2048,
-        tokenCostInMillicents: 0.04,
+        promptTokenCostInMillicents: 0.04,
+        completionTokenCostInMillicents: 0.04,
     },
 };
 export function getOpenAITextGenerationModelInformation(model) {
@@ -74,7 +91,8 @@ export function getOpenAITextGenerationModelInformation(model) {
             baseModel: model,
             isFineTuned: false,
             contextWindowSize: baseModelInformation.contextWindowSize,
-            tokenCostInMillicents: baseModelInformation.tokenCostInMillicents,
+            promptTokenCostInMillicents: baseModelInformation.promptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.completionTokenCostInMillicents,
         };
     }
     // Extract the base model from the fine-tuned model:
@@ -86,7 +104,8 @@ export function getOpenAITextGenerationModelInformation(model) {
             baseModel: baseModel,
             isFineTuned: true,
             contextWindowSize: baseModelInformation.contextWindowSize,
-            tokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+            promptTokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
         };
     }
     throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
@@ -94,8 +113,13 @@ export function getOpenAITextGenerationModelInformation(model) {
 export const isOpenAITextGenerationModel = (model) => model in OPENAI_TEXT_GENERATION_MODELS ||
     model.startsWith("ft:davinci-002:") ||
     model.startsWith("ft:babbage-002:");
-export const calculateOpenAITextGenerationCostInMillicents = ({ model, response, }) => response.usage.total_tokens *
-    getOpenAITextGenerationModelInformation(model).tokenCostInMillicents;
+export const calculateOpenAITextGenerationCostInMillicents = ({ model, response, }) => {
+    const modelInformation = getOpenAITextGenerationModelInformation(model);
+    return (response.usage.prompt_tokens *
+        modelInformation.promptTokenCostInMillicents +
+        response.usage.completion_tokens *
+            modelInformation.completionTokenCostInMillicents);
+};
 /**
  * Create a text generation model that calls the OpenAI text completion API.
  *
@@ -103,7 +127,7 @@ export const calculateOpenAITextGenerationCostInMillicents = ({ model, response,
  *
  * @example
  * const model = new OpenAITextGenerationModel({
- *   model: "text-davinci-003",
+ *   model: "gpt-3.5-turbo-instruct",
  *   temperature: 0.7,
  *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),

package/model-provider/openai/TikTokenTokenizer.cjs

@@ -57,8 +57,6 @@ function getEncodingNameForModel(model) {
         case "text-davinci-003": {
             return "p50k_base";
         }
-        case "babbage-002":
-        case "davinci-002":
         case "ada":
         case "babbage":
         case "curie":
@@ -68,11 +66,14 @@ function getEncodingNameForModel(model) {
         case "text-curie-001": {
             return "r50k_base";
         }
+        case "babbage-002":
+        case "davinci-002":
         case "gpt-3.5-turbo":
         case "gpt-3.5-turbo-0301":
         case "gpt-3.5-turbo-0613":
         case "gpt-3.5-turbo-16k":
         case "gpt-3.5-turbo-16k-0613":
+        case "gpt-3.5-turbo-instruct":
         case "gpt-4":
         case "gpt-4-0314":
         case "gpt-4-0613":

package/model-provider/openai/TikTokenTokenizer.js

@@ -53,8 +53,6 @@ function getEncodingNameForModel(model) {
         case "text-davinci-003": {
             return "p50k_base";
         }
-        case "babbage-002":
-        case "davinci-002":
         case "ada":
         case "babbage":
         case "curie":
@@ -64,11 +62,14 @@ function getEncodingNameForModel(model) {
         case "text-curie-001": {
             return "r50k_base";
         }
+        case "babbage-002":
+        case "davinci-002":
         case "gpt-3.5-turbo":
         case "gpt-3.5-turbo-0301":
         case "gpt-3.5-turbo-0613":
         case "gpt-3.5-turbo-16k":
         case "gpt-3.5-turbo-16k-0613":
+        case "gpt-3.5-turbo-instruct":
         case "gpt-4":
         case "gpt-4-0314":
         case "gpt-4-0613":

package/model-provider/openai/chat/OpenAIChatModel.cjs

@@ -6,10 +6,10 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.OpenAIChatResponseFormat = exports.OpenAIChatModel = exports.calculateOpenAIChatCostInMillicents = exports.isOpenAIChatModel = exports.getOpenAIChatModelInformation = exports.OPENAI_CHAT_MODELS = void 0;
 const secure_json_parse_1 = __importDefault(require("secure-json-parse"));
 const zod_1 = __importDefault(require("zod"));
-const AbstractModel_js_1 = require("../../../model-function/AbstractModel.cjs");
-const PromptFormatTextGenerationModel_js_1 = require("../../../prompt/PromptFormatTextGenerationModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../../core/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../../core/api/postToApi.cjs");
+const AbstractModel_js_1 = require("../../../model-function/AbstractModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../../prompt/PromptFormatTextGenerationModel.cjs");
 const OpenAIApiConfiguration_js_1 = require("../OpenAIApiConfiguration.cjs");
 const OpenAIError_js_1 = require("../OpenAIError.cjs");
 const TikTokenTokenizer_js_1 = require("../TikTokenTokenizer.cjs");
@@ -245,15 +245,21 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
      *
      * @see https://platform.openai.com/docs/guides/gpt/function-calling
      */
-    generateStructureResponse(prompt, options) {
-        const settingsWithFunctionCall = Object.assign({}, options, {
-            functionCall: prompt.functionCall,
-            functions: prompt.functions,
-        });
-        return this.callAPI(prompt.messages, {
+    generateStructureResponse(structureDefinition, prompt, options) {
+        return this.callAPI(prompt, {
             responseFormat: exports.OpenAIChatResponseFormat.json,
             functionId: options?.functionId,
-            settings: settingsWithFunctionCall,
+            settings: {
+                ...options,
+                functionCall: { name: structureDefinition.name },
+                functions: [
+                    {
+                        name: structureDefinition.name,
+                        description: structureDefinition.description,
+                        parameters: structureDefinition.schema.getJsonSchema(),
+                    },
+                ],
+            },
             run: options?.run,
         });
     }
@@ -261,6 +267,38 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
         const jsonText = response.choices[0].message.function_call.arguments;
         return secure_json_parse_1.default.parse(jsonText);
     }
+    generateStructureOrTextResponse(structureDefinitions, prompt, options) {
+        return this.callAPI(prompt, {
+            responseFormat: exports.OpenAIChatResponseFormat.json,
+            functionId: options?.functionId,
+            settings: {
+                ...options,
+                functionCall: "auto",
+                functions: structureDefinitions.map((structureDefinition) => ({
+                    name: structureDefinition.name,
+                    description: structureDefinition.description,
+                    parameters: structureDefinition.schema.getJsonSchema(),
+                })),
+            },
+            run: options?.run,
+        });
+    }
+    extractStructureAndText(response) {
+        const message = response.choices[0].message;
+        const content = message.content;
+        const functionCall = message.function_call;
+        return functionCall == null
+            ? {
+                structure: null,
+                value: null,
+                text: content ?? "",
+            }
+            : {
+                structure: functionCall.name,
+                value: secure_json_parse_1.default.parse(functionCall.arguments),
+                text: content,
+            };
+    }
     extractUsage(response) {
         return {
             promptTokens: response.usage.prompt_tokens,

package/model-provider/openai/chat/OpenAIChatModel.d.ts

@@ -1,17 +1,17 @@
 import z from "zod";
+import { ApiConfiguration } from "../../../core/api/ApiConfiguration.js";
+import { ResponseHandler } from "../../../core/api/postToApi.js";
+import { StructureDefinition } from "../../../core/structure/StructureDefinition.js";
 import { AbstractModel } from "../../../model-function/AbstractModel.js";
 import { ModelFunctionOptions } from "../../../model-function/ModelFunctionOptions.js";
-import { ApiConfiguration } from "../../../core/api/ApiConfiguration.js";
 import { StructureGenerationModel } from "../../../model-function/generate-structure/StructureGenerationModel.js";
 import { StructureOrTextGenerationModel } from "../../../model-function/generate-structure/StructureOrTextGenerationModel.js";
 import { DeltaEvent } from "../../../model-function/generate-text/DeltaEvent.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../../model-function/generate-text/TextGenerationModel.js";
 import { PromptFormat } from "../../../prompt/PromptFormat.js";
 import { PromptFormatTextGenerationModel } from "../../../prompt/PromptFormatTextGenerationModel.js";
-import { ResponseHandler } from "../../../core/api/postToApi.js";
 import { TikTokenTokenizer } from "../TikTokenTokenizer.js";
 import { OpenAIChatMessage } from "./OpenAIChatMessage.js";
-import { OpenAIChatAutoFunctionPrompt, OpenAIChatSingleFunctionPrompt, OpenAIFunctionDescription } from "./OpenAIChatPrompt.js";
 import { OpenAIChatDelta } from "./OpenAIChatStreamIterable.js";
 export declare const OPENAI_CHAT_MODELS: {
     "gpt-4": {
@@ -132,7 +132,7 @@ export interface OpenAIChatSettings extends TextGenerationModelSettings, Omit<Op
  *   ),
  * ]);
  */
-export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> implements TextGenerationModel<OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings>, StructureGenerationModel<OpenAIChatSingleFunctionPrompt<unknown>, OpenAIChatResponse, OpenAIChatSettings>, StructureOrTextGenerationModel<OpenAIChatAutoFunctionPrompt<Array<OpenAIFunctionDescription<unknown>>>, OpenAIChatResponse, OpenAIChatSettings> {
+export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> implements TextGenerationModel<OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings>, StructureGenerationModel<OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatSettings>, StructureOrTextGenerationModel<OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatSettings> {
     constructor(settings: OpenAIChatSettings);
     readonly provider: "openai";
     get modelName(): OpenAIChatModelType;
@@ -183,8 +183,18 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
      *
      * @see https://platform.openai.com/docs/guides/gpt/function-calling
      */
-    generateStructureResponse(prompt: OpenAIChatSingleFunctionPrompt<unknown> | OpenAIChatAutoFunctionPrompt<Array<OpenAIFunctionDescription<unknown>>>, options?: ModelFunctionOptions<OpenAIChatSettings> | undefined): PromiseLike<OpenAIChatResponse>;
+    generateStructureResponse(structureDefinition: StructureDefinition<string, unknown>, prompt: OpenAIChatMessage[], options?: ModelFunctionOptions<OpenAIChatSettings> | undefined): PromiseLike<OpenAIChatResponse>;
     extractStructure(response: OpenAIChatResponse): unknown;
+    generateStructureOrTextResponse(structureDefinitions: Array<StructureDefinition<string, unknown>>, prompt: OpenAIChatMessage[], options?: ModelFunctionOptions<OpenAIChatSettings> | undefined): PromiseLike<OpenAIChatResponse>;
+    extractStructureAndText(response: OpenAIChatResponse): {
+        structure: null;
+        value: null;
+        text: string;
+    } | {
+        structure: string;
+        value: any;
+        text: string | null;
+    };
     extractUsage(response: OpenAIChatResponse): {
         promptTokens: number;
         completionTokens: number;

package/model-provider/openai/chat/OpenAIChatModel.js

@@ -1,9 +1,9 @@
 import SecureJSON from "secure-json-parse";
 import z from "zod";
-import { AbstractModel } from "../../../model-function/AbstractModel.js";
-import { PromptFormatTextGenerationModel } from "../../../prompt/PromptFormatTextGenerationModel.js";
 import { callWithRetryAndThrottle } from "../../../core/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../../core/api/postToApi.js";
+import { AbstractModel } from "../../../model-function/AbstractModel.js";
+import { PromptFormatTextGenerationModel } from "../../../prompt/PromptFormatTextGenerationModel.js";
 import { OpenAIApiConfiguration } from "../OpenAIApiConfiguration.js";
 import { failedOpenAICallResponseHandler } from "../OpenAIError.js";
 import { TikTokenTokenizer } from "../TikTokenTokenizer.js";
@@ -236,15 +236,21 @@ export class OpenAIChatModel extends AbstractModel {
      *
      * @see https://platform.openai.com/docs/guides/gpt/function-calling
      */
-    generateStructureResponse(prompt, options) {
-        const settingsWithFunctionCall = Object.assign({}, options, {
-            functionCall: prompt.functionCall,
-            functions: prompt.functions,
-        });
-        return this.callAPI(prompt.messages, {
+    generateStructureResponse(structureDefinition, prompt, options) {
+        return this.callAPI(prompt, {
             responseFormat: OpenAIChatResponseFormat.json,
             functionId: options?.functionId,
-            settings: settingsWithFunctionCall,
+            settings: {
+                ...options,
+                functionCall: { name: structureDefinition.name },
+                functions: [
+                    {
+                        name: structureDefinition.name,
+                        description: structureDefinition.description,
+                        parameters: structureDefinition.schema.getJsonSchema(),
+                    },
+                ],
+            },
             run: options?.run,
         });
     }
@@ -252,6 +258,38 @@ export class OpenAIChatModel extends AbstractModel {
         const jsonText = response.choices[0].message.function_call.arguments;
         return SecureJSON.parse(jsonText);
     }
+    generateStructureOrTextResponse(structureDefinitions, prompt, options) {
+        return this.callAPI(prompt, {
+            responseFormat: OpenAIChatResponseFormat.json,
+            functionId: options?.functionId,
+            settings: {
+                ...options,
+                functionCall: "auto",
+                functions: structureDefinitions.map((structureDefinition) => ({
+                    name: structureDefinition.name,
+                    description: structureDefinition.description,
+                    parameters: structureDefinition.schema.getJsonSchema(),
+                })),
+            },
+            run: options?.run,
+        });
+    }
+    extractStructureAndText(response) {
+        const message = response.choices[0].message;
+        const content = message.content;
+        const functionCall = message.function_call;
+        return functionCall == null
+            ? {
+                structure: null,
+                value: null,
+                text: content ?? "",
+            }
+            : {
+                structure: functionCall.name,
+                value: SecureJSON.parse(functionCall.arguments),
+                text: content,
+            };
+    }
     extractUsage(response) {
         return {
             promptTokens: response.usage.prompt_tokens,