modelfusion 0.23.0 → 0.24.1

package/README.md

@@ -35,6 +35,8 @@ You need to install `zod` and a matching version of `zod-to-json-schema` (peer d
 npm install zod zod-to-json-schema
 ```
 
+Or use a template: [ModelFusion terminal app starter](https://github.com/lgrammel/modelfusion-terminal-app-starter)
+
 ## Usage Examples
 
 You can provide API keys for the different [integrations](https://modelfusion.dev/integration/model-provider/) using environment variables (e.g., `OPENAI_API_KEY`) or pass them into the model constructors as options.

package/model-provider/cohere/CohereTokenizer.d.ts

@@ -4,7 +4,7 @@ import { Run } from "../../core/Run.js";
 import { RetryFunction } from "../../util/api/RetryFunction.js";
 import { ThrottleFunction } from "../../util/api/ThrottleFunction.js";
 import { CohereTextGenerationModelType } from "./CohereTextGenerationModel.js";
-import { CohereTextEmbeddingModelType } from "./index.js";
+import { CohereTextEmbeddingModelType } from "./CohereTextEmbeddingModel.js";
 export type CohereTokenizerModelType = CohereTextGenerationModelType | CohereTextEmbeddingModelType;
 export interface CohereTokenizerSettings {
     model: CohereTokenizerModelType;

package/model-provider/openai/OpenAITextGenerationModel.cjs

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.OpenAITextResponseFormat = exports.OpenAITextGenerationModel = exports.calculateOpenAITextGenerationCostInMillicents = exports.isOpenAITextGenerationModel = exports.OPENAI_TEXT_GENERATION_MODELS = void 0;
+exports.OpenAITextResponseFormat = exports.OpenAITextGenerationModel = exports.calculateOpenAITextGenerationCostInMillicents = exports.isOpenAITextGenerationModel = exports.getOpenAITextGenerationModelInformation = exports.OPENAI_TEXT_GENERATION_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");
@@ -23,10 +23,12 @@ exports.OPENAI_TEXT_GENERATION_MODELS = {
     "davinci-002": {
         contextWindowSize: 16384,
         tokenCostInMillicents: 0.2,
+        fineTunedTokenCostInMillicents: 1.2,
     },
     "babbage-002": {
         contextWindowSize: 16384,
         tokenCostInMillicents: 0.04,
+        fineTunedTokenCostInMillicents: 0.16,
     },
     "text-davinci-003": {
         contextWindowSize: 4096,
@@ -69,10 +71,38 @@ exports.OPENAI_TEXT_GENERATION_MODELS = {
         tokenCostInMillicents: 0.04,
     },
 };
-const isOpenAITextGenerationModel = (model) => model in exports.OPENAI_TEXT_GENERATION_MODELS;
+function getOpenAITextGenerationModelInformation(model) {
+    // Model is already a base model:
+    if (model in exports.OPENAI_TEXT_GENERATION_MODELS) {
+        const baseModelInformation = exports.OPENAI_TEXT_GENERATION_MODELS[model];
+        return {
+            baseModel: model,
+            isFineTuned: false,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            tokenCostInMillicents: baseModelInformation.tokenCostInMillicents,
+        };
+    }
+    // Extract the base model from the fine-tuned model:
+    // eslint-disable-next-line @typescript-eslint/no-unused-vars
+    const [_, baseModel, ___, ____, _____] = model.split(":");
+    if (["davinci-002", "babbage-002"].includes(baseModel)) {
+        const baseModelInformation = exports.OPENAI_TEXT_GENERATION_MODELS[baseModel];
+        return {
+            baseModel: baseModel,
+            isFineTuned: true,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            tokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+        };
+    }
+    throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
+}
+exports.getOpenAITextGenerationModelInformation = getOpenAITextGenerationModelInformation;
+const isOpenAITextGenerationModel = (model) => model in exports.OPENAI_TEXT_GENERATION_MODELS ||
+    model.startsWith("ft:davinci-002:") ||
+    model.startsWith("ft:babbage-002:");
 exports.isOpenAITextGenerationModel = isOpenAITextGenerationModel;
 const calculateOpenAITextGenerationCostInMillicents = ({ model, response, }) => response.usage.total_tokens *
-    exports.OPENAI_TEXT_GENERATION_MODELS[model].tokenCostInMillicents;
+    getOpenAITextGenerationModelInformation(model).tokenCostInMillicents;
 exports.calculateOpenAITextGenerationCostInMillicents = calculateOpenAITextGenerationCostInMillicents;
 /**
  * Create a text generation model that calls the OpenAI text completion API.
@@ -113,9 +143,11 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
             writable: true,
             value: void 0
         });
-        this.tokenizer = new TikTokenTokenizer_js_1.TikTokenTokenizer({ model: settings.model });
-        this.contextWindowSize =
-            exports.OPENAI_TEXT_GENERATION_MODELS[settings.model].contextWindowSize;
+        const modelInformation = getOpenAITextGenerationModelInformation(this.settings.model);
+        this.tokenizer = new TikTokenTokenizer_js_1.TikTokenTokenizer({
+            model: modelInformation.baseModel,
+        });
+        this.contextWindowSize = modelInformation.contextWindowSize;
     }
     get modelName() {
         return this.settings.model;
@@ -169,6 +201,7 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
             "presencePenalty",
             "frequencyPenalty",
             "bestOf",
+            "logitBias",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -243,7 +276,7 @@ const openAITextGenerationResponseSchema = zod_1.default.object({
  *
  * console.log(response.choices[0].text);
  */
-async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, user, }) {
+async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, logitBias, user, }) {
     return (0, postToApi_js_1.postJsonToApi)({
         url: `${baseUrl}/completions`,
         headers: {
@@ -265,6 +298,7 @@ async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
             best_of: bestOf,
+            logit_bias: logitBias,
             user,
         },
         failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,

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

@@ -5,8 +5,6 @@ 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 { RetryFunction } from "../../util/api/RetryFunction.js";
-import { ThrottleFunction } from "../../util/api/ThrottleFunction.js";
 import { ResponseHandler } from "../../util/api/postToApi.js";
 import { OpenAIImageGenerationCallSettings } from "./OpenAIImageGenerationModel.js";
 import { OpenAIModelSettings } from "./OpenAIModelSettings.js";
@@ -19,10 +17,12 @@ export declare const OPENAI_TEXT_GENERATION_MODELS: {
     "davinci-002": {
         contextWindowSize: number;
         tokenCostInMillicents: number;
+        fineTunedTokenCostInMillicents: number;
     };
     "babbage-002": {
         contextWindowSize: number;
         tokenCostInMillicents: number;
+        fineTunedTokenCostInMillicents: number;
     };
     "text-davinci-003": {
         contextWindowSize: number;
@@ -65,29 +65,39 @@ export declare const OPENAI_TEXT_GENERATION_MODELS: {
         tokenCostInMillicents: number;
     };
 };
-export type OpenAITextGenerationModelType = keyof typeof OPENAI_TEXT_GENERATION_MODELS;
-export declare const isOpenAITextGenerationModel: (model: string) => model is "davinci-002" | "babbage-002" | "text-davinci-003" | "text-davinci-002" | "code-davinci-002" | "davinci" | "text-curie-001" | "curie" | "text-babbage-001" | "babbage" | "text-ada-001" | "ada";
+export declare function getOpenAITextGenerationModelInformation(model: OpenAITextGenerationModelType): {
+    baseModel: OpenAITextGenerationBaseModelType;
+    isFineTuned: boolean;
+    contextWindowSize: number;
+    tokenCostInMillicents: number;
+};
+type FineTuneableOpenAITextGenerationModelType = "davinci-002" | "babbage-002";
+type FineTunedOpenAITextGenerationModelType = `ft:${FineTuneableOpenAITextGenerationModelType}:${string}:${string}:${string}`;
+export type OpenAITextGenerationBaseModelType = keyof typeof OPENAI_TEXT_GENERATION_MODELS;
+export type OpenAITextGenerationModelType = OpenAITextGenerationBaseModelType | FineTunedOpenAITextGenerationModelType;
+export declare const isOpenAITextGenerationModel: (model: string) => model is OpenAITextGenerationModelType;
 export declare const calculateOpenAITextGenerationCostInMillicents: ({ model, response, }: {
     model: OpenAITextGenerationModelType;
     response: OpenAITextGenerationResponse;
 }) => number;
-export interface OpenAITextGenerationModelSettings extends TextGenerationModelSettings {
+export interface OpenAITextGenerationCallSettings {
     model: OpenAITextGenerationModelType;
     headers?: Record<string, string>;
-    baseUrl?: string;
-    apiKey?: string;
-    retry?: RetryFunction;
-    throttle?: ThrottleFunction;
-    isUserIdForwardingEnabled?: boolean;
     suffix?: string;
+    maxTokens?: number;
     temperature?: number;
     topP?: number;
     n?: number;
     logprobs?: number;
     echo?: boolean;
+    stop?: string | string[];
     presencePenalty?: number;
     frequencyPenalty?: number;
     bestOf?: number;
+    logitBias?: Record<number, number>;
+}
+export interface OpenAITextGenerationModelSettings extends TextGenerationModelSettings, OpenAIModelSettings, Omit<OpenAITextGenerationCallSettings, "stop" | "maxTokens"> {
+    isUserIdForwardingEnabled?: boolean;
 }
 /**
  * Create a text generation model that calls the OpenAI text completion API.
@@ -110,7 +120,7 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
 export declare class OpenAITextGenerationModel extends AbstractModel<OpenAITextGenerationModelSettings> implements TextGenerationModel<string, OpenAITextGenerationResponse, OpenAITextGenerationDelta, OpenAITextGenerationModelSettings> {
     constructor(settings: OpenAITextGenerationModelSettings);
     readonly provider: "openai";
-    get modelName(): "davinci-002" | "babbage-002" | "text-davinci-003" | "text-davinci-002" | "code-davinci-002" | "davinci" | "text-curie-001" | "curie" | "text-babbage-001" | "babbage" | "text-ada-001" | "ada";
+    get modelName(): OpenAITextGenerationModelType;
     readonly contextWindowSize: number;
     readonly tokenizer: TikTokenTokenizer;
     private get apiKey();
@@ -126,8 +136,8 @@ export declare class OpenAITextGenerationModel extends AbstractModel<OpenAITextG
         model: string;
         usage: {
             prompt_tokens: number;
-            total_tokens: number;
             completion_tokens: number;
+            total_tokens: number;
         };
         id: string;
         created: number;
@@ -176,20 +186,20 @@ declare const openAITextGenerationResponseSchema: z.ZodObject<{
         total_tokens: z.ZodNumber;
     }, "strip", z.ZodTypeAny, {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     }, {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     }>;
 }, "strip", z.ZodTypeAny, {
     object: "text_completion";
     model: string;
     usage: {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     };
     id: string;
     created: number;
@@ -204,8 +214,8 @@ declare const openAITextGenerationResponseSchema: z.ZodObject<{
     model: string;
     usage: {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     };
     id: string;
     created: number;
@@ -232,8 +242,8 @@ export declare const OpenAITextResponseFormat: {
             model: string;
             usage: {
                 prompt_tokens: number;
-                total_tokens: number;
                 completion_tokens: number;
+                total_tokens: number;
             };
             id: string;
             created: number;

package/model-provider/openai/OpenAITextGenerationModel.js

@@ -17,10 +17,12 @@ export const OPENAI_TEXT_GENERATION_MODELS = {
     "davinci-002": {
         contextWindowSize: 16384,
         tokenCostInMillicents: 0.2,
+        fineTunedTokenCostInMillicents: 1.2,
     },
     "babbage-002": {
         contextWindowSize: 16384,
         tokenCostInMillicents: 0.04,
+        fineTunedTokenCostInMillicents: 0.16,
     },
     "text-davinci-003": {
         contextWindowSize: 4096,
@@ -63,9 +65,36 @@ export const OPENAI_TEXT_GENERATION_MODELS = {
         tokenCostInMillicents: 0.04,
     },
 };
-export const isOpenAITextGenerationModel = (model) => model in OPENAI_TEXT_GENERATION_MODELS;
+export function getOpenAITextGenerationModelInformation(model) {
+    // Model is already a base model:
+    if (model in OPENAI_TEXT_GENERATION_MODELS) {
+        const baseModelInformation = OPENAI_TEXT_GENERATION_MODELS[model];
+        return {
+            baseModel: model,
+            isFineTuned: false,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            tokenCostInMillicents: baseModelInformation.tokenCostInMillicents,
+        };
+    }
+    // Extract the base model from the fine-tuned model:
+    // eslint-disable-next-line @typescript-eslint/no-unused-vars
+    const [_, baseModel, ___, ____, _____] = model.split(":");
+    if (["davinci-002", "babbage-002"].includes(baseModel)) {
+        const baseModelInformation = OPENAI_TEXT_GENERATION_MODELS[baseModel];
+        return {
+            baseModel: baseModel,
+            isFineTuned: true,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            tokenCostInMillicents: baseModelInformation.fineTunedTokenCostInMillicents,
+        };
+    }
+    throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
+}
+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 *
-    OPENAI_TEXT_GENERATION_MODELS[model].tokenCostInMillicents;
+    getOpenAITextGenerationModelInformation(model).tokenCostInMillicents;
 /**
  * Create a text generation model that calls the OpenAI text completion API.
  *
@@ -105,9 +134,11 @@ export class OpenAITextGenerationModel extends AbstractModel {
             writable: true,
             value: void 0
         });
-        this.tokenizer = new TikTokenTokenizer({ model: settings.model });
-        this.contextWindowSize =
-            OPENAI_TEXT_GENERATION_MODELS[settings.model].contextWindowSize;
+        const modelInformation = getOpenAITextGenerationModelInformation(this.settings.model);
+        this.tokenizer = new TikTokenTokenizer({
+            model: modelInformation.baseModel,
+        });
+        this.contextWindowSize = modelInformation.contextWindowSize;
     }
     get modelName() {
         return this.settings.model;
@@ -161,6 +192,7 @@ export class OpenAITextGenerationModel extends AbstractModel {
             "presencePenalty",
             "frequencyPenalty",
             "bestOf",
+            "logitBias",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -234,7 +266,7 @@ const openAITextGenerationResponseSchema = z.object({
  *
  * console.log(response.choices[0].text);
  */
-async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, user, }) {
+async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, prompt, suffix, maxTokens, temperature, topP, n, logprobs, echo, stop, presencePenalty, frequencyPenalty, bestOf, logitBias, user, }) {
     return postJsonToApi({
         url: `${baseUrl}/completions`,
         headers: {
@@ -256,6 +288,7 @@ async function callOpenAITextGenerationAPI({ baseUrl = "https://api.openai.com/v
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
             best_of: bestOf,
+            logit_bias: logitBias,
             user,
         },
         failedResponseHandler: failedOpenAICallResponseHandler,

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

@@ -1,6 +1,8 @@
 import { TiktokenEncoding } from "js-tiktoken";
 import { FullTokenizer } from "../../model-function/tokenize-text/Tokenizer.js";
-import { OpenAIChatModelType, OpenAITextEmbeddingModelType, OpenAITextGenerationModelType } from "./index.js";
+import { OpenAITextEmbeddingModelType } from "./OpenAITextEmbeddingModel.js";
+import { OpenAITextGenerationBaseModelType } from "./OpenAITextGenerationModel.js";
+import { OpenAIChatBaseModelType } from "./chat/OpenAIChatModel.js";
 /**
  * TikToken tokenizer for OpenAI language models.
  *
@@ -21,7 +23,7 @@ export declare class TikTokenTokenizer implements FullTokenizer {
      * Get a TikToken tokenizer for a specific model or encoding.
      */
     constructor(options: {
-        model: OpenAIChatModelType | OpenAITextEmbeddingModelType | OpenAITextGenerationModelType;
+        model: OpenAIChatBaseModelType | OpenAITextGenerationBaseModelType | OpenAITextEmbeddingModelType;
     } | {
         encoding: TiktokenEncoding;
     });

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

@@ -3,7 +3,7 @@ var __importDefault = (this && this.__importDefault) || function (mod) {
     return (mod && mod.__esModule) ? mod : { "default": mod };
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.OpenAIChatResponseFormat = exports.OpenAIChatModel = exports.calculateOpenAIChatCostInMillicents = exports.isOpenAIChatModel = exports.OPENAI_CHAT_MODELS = void 0;
+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");
@@ -55,6 +55,8 @@ exports.OPENAI_CHAT_MODELS = {
         contextWindowSize: 4096,
         promptTokenCostInMillicents: 0.15,
         completionTokenCostInMillicents: 0.2,
+        fineTunedPromptTokenCostInMillicents: 1.2,
+        fineTunedCompletionTokenCostInMillicents: 1.6,
     },
     "gpt-3.5-turbo-0301": {
         contextWindowSize: 4096,
@@ -65,6 +67,8 @@ exports.OPENAI_CHAT_MODELS = {
         contextWindowSize: 4096,
         promptTokenCostInMillicents: 0.15,
         completionTokenCostInMillicents: 0.2,
+        fineTunedPromptTokenCostInMillicents: 1.2,
+        fineTunedCompletionTokenCostInMillicents: 1.6,
     },
     "gpt-3.5-turbo-16k": {
         contextWindowSize: 16384,
@@ -77,12 +81,45 @@ exports.OPENAI_CHAT_MODELS = {
         completionTokenCostInMillicents: 0.4,
     },
 };
-const isOpenAIChatModel = (model) => model in exports.OPENAI_CHAT_MODELS;
+function getOpenAIChatModelInformation(model) {
+    // Model is already a base model:
+    if (model in exports.OPENAI_CHAT_MODELS) {
+        const baseModelInformation = exports.OPENAI_CHAT_MODELS[model];
+        return {
+            baseModel: model,
+            isFineTuned: false,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            promptTokenCostInMillicents: baseModelInformation.promptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.completionTokenCostInMillicents,
+        };
+    }
+    // Extract the base model from the fine-tuned model:
+    // eslint-disable-next-line @typescript-eslint/no-unused-vars
+    const [_, baseModel, ___, ____, _____] = model.split(":");
+    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613"].includes(baseModel)) {
+        const baseModelInformation = exports.OPENAI_CHAT_MODELS[baseModel];
+        return {
+            baseModel: baseModel,
+            isFineTuned: true,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            promptTokenCostInMillicents: baseModelInformation.fineTunedPromptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.fineTunedCompletionTokenCostInMillicents,
+        };
+    }
+    throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
+}
+exports.getOpenAIChatModelInformation = getOpenAIChatModelInformation;
+const isOpenAIChatModel = (model) => model in exports.OPENAI_CHAT_MODELS ||
+    model.startsWith("ft:gpt-3.5-turbo-0613:") ||
+    model.startsWith("ft:gpt-3.5-turbo:");
 exports.isOpenAIChatModel = isOpenAIChatModel;
-const calculateOpenAIChatCostInMillicents = ({ model, response, }) => response.usage.prompt_tokens *
-    exports.OPENAI_CHAT_MODELS[model].promptTokenCostInMillicents +
-    response.usage.completion_tokens *
-        exports.OPENAI_CHAT_MODELS[model].completionTokenCostInMillicents;
+const calculateOpenAIChatCostInMillicents = ({ model, response, }) => {
+    const modelInformation = getOpenAIChatModelInformation(model);
+    return (response.usage.prompt_tokens *
+        modelInformation.promptTokenCostInMillicents +
+        response.usage.completion_tokens *
+            modelInformation.completionTokenCostInMillicents);
+};
 exports.calculateOpenAIChatCostInMillicents = calculateOpenAIChatCostInMillicents;
 /**
  * Create a text generation model that calls the OpenAI chat completion API.
@@ -124,9 +161,11 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
             writable: true,
             value: void 0
         });
-        this.tokenizer = new TikTokenTokenizer_js_1.TikTokenTokenizer({ model: this.settings.model });
-        this.contextWindowSize =
-            exports.OPENAI_CHAT_MODELS[this.settings.model].contextWindowSize;
+        const modelInformation = getOpenAIChatModelInformation(this.settings.model);
+        this.tokenizer = new TikTokenTokenizer_js_1.TikTokenTokenizer({
+            model: modelInformation.baseModel,
+        });
+        this.contextWindowSize = modelInformation.contextWindowSize;
     }
     get modelName() {
         return this.settings.model;
@@ -180,6 +219,9 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
             "temperature",
             "topP",
             "n",
+            "presencePenalty",
+            "frequencyPenalty",
+            "logitBias",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -268,7 +310,7 @@ const openAIChatResponseSchema = zod_1.default.object({
         total_tokens: zod_1.default.number(),
     }),
 });
-async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, messages, functions, functionCall, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, user, }) {
+async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, messages, functions, functionCall, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, logitBias, user, }) {
     return (0, postToApi_js_1.postJsonToApi)({
         url: `${baseUrl}/chat/completions`,
         headers: {
@@ -288,6 +330,7 @@ async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v
             max_tokens: maxTokens,
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
+            logit_bias: logitBias,
             user,
         },
         failedResponseHandler: OpenAIError_js_1.failedOpenAICallResponseHandler,

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

@@ -48,6 +48,8 @@ export declare const OPENAI_CHAT_MODELS: {
         contextWindowSize: number;
         promptTokenCostInMillicents: number;
         completionTokenCostInMillicents: number;
+        fineTunedPromptTokenCostInMillicents: number;
+        fineTunedCompletionTokenCostInMillicents: number;
     };
     "gpt-3.5-turbo-0301": {
         contextWindowSize: number;
@@ -58,6 +60,8 @@ export declare const OPENAI_CHAT_MODELS: {
         contextWindowSize: number;
         promptTokenCostInMillicents: number;
         completionTokenCostInMillicents: number;
+        fineTunedPromptTokenCostInMillicents: number;
+        fineTunedCompletionTokenCostInMillicents: number;
     };
     "gpt-3.5-turbo-16k": {
         contextWindowSize: number;
@@ -70,8 +74,18 @@ export declare const OPENAI_CHAT_MODELS: {
         completionTokenCostInMillicents: number;
     };
 };
-export type OpenAIChatModelType = keyof typeof OPENAI_CHAT_MODELS;
-export declare const isOpenAIChatModel: (model: string) => model is "gpt-4" | "gpt-4-0314" | "gpt-4-0613" | "gpt-4-32k" | "gpt-4-32k-0314" | "gpt-4-32k-0613" | "gpt-3.5-turbo" | "gpt-3.5-turbo-0301" | "gpt-3.5-turbo-0613" | "gpt-3.5-turbo-16k" | "gpt-3.5-turbo-16k-0613";
+export declare function getOpenAIChatModelInformation(model: OpenAIChatModelType): {
+    baseModel: OpenAIChatBaseModelType;
+    isFineTuned: boolean;
+    contextWindowSize: number;
+    promptTokenCostInMillicents: number;
+    completionTokenCostInMillicents: number;
+};
+type FineTuneableOpenAIChatModelType = `gpt-3.5-turbo` | `gpt-3.5-turbo-0613`;
+type FineTunedOpenAIChatModelType = `ft:${FineTuneableOpenAIChatModelType}:${string}:${string}:${string}`;
+export type OpenAIChatBaseModelType = keyof typeof OPENAI_CHAT_MODELS;
+export type OpenAIChatModelType = OpenAIChatBaseModelType | FineTunedOpenAIChatModelType;
+export declare const isOpenAIChatModel: (model: string) => model is OpenAIChatModelType;
 export declare const calculateOpenAIChatCostInMillicents: ({ model, response, }: {
     model: OpenAIChatModelType;
     response: OpenAIChatResponse;
@@ -87,13 +101,14 @@ export interface OpenAIChatCallSettings {
     functionCall?: "none" | "auto" | {
         name: string;
     };
+    stop?: string | string[];
+    maxTokens?: number;
     temperature?: number;
     topP?: number;
     n?: number;
-    stop?: string | string[];
-    maxTokens?: number;
     presencePenalty?: number;
     frequencyPenalty?: number;
+    logitBias?: Record<number, number>;
 }
 export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIModelSettings, Omit<OpenAIChatCallSettings, "stop" | "maxTokens"> {
     isUserIdForwardingEnabled?: boolean;
@@ -120,7 +135,7 @@ export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIM
 export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> implements TextGenerationModel<OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings>, JsonGenerationModel<OpenAIChatSingleFunctionPrompt<unknown>, OpenAIChatResponse, OpenAIChatSettings>, JsonOrTextGenerationModel<OpenAIChatAutoFunctionPrompt<Array<OpenAIFunctionDescription<unknown>>>, OpenAIChatResponse, OpenAIChatSettings> {
     constructor(settings: OpenAIChatSettings);
     readonly provider: "openai";
-    get modelName(): "gpt-4" | "gpt-4-0314" | "gpt-4-0613" | "gpt-4-32k" | "gpt-4-32k-0314" | "gpt-4-32k-0613" | "gpt-3.5-turbo" | "gpt-3.5-turbo-0301" | "gpt-3.5-turbo-0613" | "gpt-3.5-turbo-16k" | "gpt-3.5-turbo-16k-0613";
+    get modelName(): OpenAIChatModelType;
     readonly contextWindowSize: number;
     readonly tokenizer: TikTokenTokenizer;
     private get apiKey();
@@ -140,8 +155,8 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
         model: string;
         usage: {
             prompt_tokens: number;
-            total_tokens: number;
             completion_tokens: number;
+            total_tokens: number;
         };
         id: string;
         created: number;
@@ -247,20 +262,20 @@ declare const openAIChatResponseSchema: z.ZodObject<{
         total_tokens: z.ZodNumber;
     }, "strip", z.ZodTypeAny, {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     }, {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     }>;
 }, "strip", z.ZodTypeAny, {
     object: "chat.completion";
     model: string;
     usage: {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     };
     id: string;
     created: number;
@@ -282,8 +297,8 @@ declare const openAIChatResponseSchema: z.ZodObject<{
     model: string;
     usage: {
         prompt_tokens: number;
-        total_tokens: number;
         completion_tokens: number;
+        total_tokens: number;
     };
     id: string;
     created: number;
@@ -317,8 +332,8 @@ export declare const OpenAIChatResponseFormat: {
             model: string;
             usage: {
                 prompt_tokens: number;
-                total_tokens: number;
                 completion_tokens: number;
+                total_tokens: number;
             };
             id: string;
             created: number;

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

@@ -49,6 +49,8 @@ export const OPENAI_CHAT_MODELS = {
         contextWindowSize: 4096,
         promptTokenCostInMillicents: 0.15,
         completionTokenCostInMillicents: 0.2,
+        fineTunedPromptTokenCostInMillicents: 1.2,
+        fineTunedCompletionTokenCostInMillicents: 1.6,
     },
     "gpt-3.5-turbo-0301": {
         contextWindowSize: 4096,
@@ -59,6 +61,8 @@ export const OPENAI_CHAT_MODELS = {
         contextWindowSize: 4096,
         promptTokenCostInMillicents: 0.15,
         completionTokenCostInMillicents: 0.2,
+        fineTunedPromptTokenCostInMillicents: 1.2,
+        fineTunedCompletionTokenCostInMillicents: 1.6,
     },
     "gpt-3.5-turbo-16k": {
         contextWindowSize: 16384,
@@ -71,11 +75,43 @@ export const OPENAI_CHAT_MODELS = {
         completionTokenCostInMillicents: 0.4,
     },
 };
-export const isOpenAIChatModel = (model) => model in OPENAI_CHAT_MODELS;
-export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => response.usage.prompt_tokens *
-    OPENAI_CHAT_MODELS[model].promptTokenCostInMillicents +
-    response.usage.completion_tokens *
-        OPENAI_CHAT_MODELS[model].completionTokenCostInMillicents;
+export function getOpenAIChatModelInformation(model) {
+    // Model is already a base model:
+    if (model in OPENAI_CHAT_MODELS) {
+        const baseModelInformation = OPENAI_CHAT_MODELS[model];
+        return {
+            baseModel: model,
+            isFineTuned: false,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            promptTokenCostInMillicents: baseModelInformation.promptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.completionTokenCostInMillicents,
+        };
+    }
+    // Extract the base model from the fine-tuned model:
+    // eslint-disable-next-line @typescript-eslint/no-unused-vars
+    const [_, baseModel, ___, ____, _____] = model.split(":");
+    if (["gpt-3.5-turbo", "gpt-3.5-turbo-0613"].includes(baseModel)) {
+        const baseModelInformation = OPENAI_CHAT_MODELS[baseModel];
+        return {
+            baseModel: baseModel,
+            isFineTuned: true,
+            contextWindowSize: baseModelInformation.contextWindowSize,
+            promptTokenCostInMillicents: baseModelInformation.fineTunedPromptTokenCostInMillicents,
+            completionTokenCostInMillicents: baseModelInformation.fineTunedCompletionTokenCostInMillicents,
+        };
+    }
+    throw new Error(`Unknown OpenAI chat base model ${baseModel}.`);
+}
+export const isOpenAIChatModel = (model) => model in OPENAI_CHAT_MODELS ||
+    model.startsWith("ft:gpt-3.5-turbo-0613:") ||
+    model.startsWith("ft:gpt-3.5-turbo:");
+export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => {
+    const modelInformation = getOpenAIChatModelInformation(model);
+    return (response.usage.prompt_tokens *
+        modelInformation.promptTokenCostInMillicents +
+        response.usage.completion_tokens *
+            modelInformation.completionTokenCostInMillicents);
+};
 /**
  * Create a text generation model that calls the OpenAI chat completion API.
  *
@@ -116,9 +152,11 @@ export class OpenAIChatModel extends AbstractModel {
             writable: true,
             value: void 0
         });
-        this.tokenizer = new TikTokenTokenizer({ model: this.settings.model });
-        this.contextWindowSize =
-            OPENAI_CHAT_MODELS[this.settings.model].contextWindowSize;
+        const modelInformation = getOpenAIChatModelInformation(this.settings.model);
+        this.tokenizer = new TikTokenTokenizer({
+            model: modelInformation.baseModel,
+        });
+        this.contextWindowSize = modelInformation.contextWindowSize;
     }
     get modelName() {
         return this.settings.model;
@@ -172,6 +210,9 @@ export class OpenAIChatModel extends AbstractModel {
             "temperature",
             "topP",
             "n",
+            "presencePenalty",
+            "frequencyPenalty",
+            "logitBias",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -259,7 +300,7 @@ const openAIChatResponseSchema = z.object({
         total_tokens: z.number(),
     }),
 });
-async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, messages, functions, functionCall, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, user, }) {
+async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v1", headers, abortSignal, responseFormat, apiKey, model, messages, functions, functionCall, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, logitBias, user, }) {
     return postJsonToApi({
         url: `${baseUrl}/chat/completions`,
         headers: {
@@ -279,6 +320,7 @@ async function callOpenAIChatCompletionAPI({ baseUrl = "https://api.openai.com/v
             max_tokens: maxTokens,
             presence_penalty: presencePenalty,
             frequency_penalty: frequencyPenalty,
+            logit_bias: logitBias,
             user,
         },
         failedResponseHandler: failedOpenAICallResponseHandler,

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

@@ -3,6 +3,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.countOpenAIChatPromptTokens = exports.countOpenAIChatMessageTokens = exports.OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT = exports.OPENAI_CHAT_PROMPT_BASE_TOKEN_COUNT = void 0;
 const countTokens_js_1 = require("../../../model-function/tokenize-text/countTokens.cjs");
 const TikTokenTokenizer_js_1 = require("../TikTokenTokenizer.cjs");
+const OpenAIChatModel_js_1 = require("./OpenAIChatModel.cjs");
 /**
  * Prompt tokens that are included automatically for every full
  * chat prompt (several messages) that is sent to OpenAI.
@@ -14,8 +15,10 @@ exports.OPENAI_CHAT_PROMPT_BASE_TOKEN_COUNT = 2;
  */
 exports.OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT = 5;
 async function countOpenAIChatMessageTokens({ message, model, }) {
-    return (exports.OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT +
-        (await (0, countTokens_js_1.countTokens)(new TikTokenTokenizer_js_1.TikTokenTokenizer({ model }), message.content ?? "")));
+    const contentTokenCount = await (0, countTokens_js_1.countTokens)(new TikTokenTokenizer_js_1.TikTokenTokenizer({
+        model: (0, OpenAIChatModel_js_1.getOpenAIChatModelInformation)(model).baseModel,
+    }), message.content ?? "");
+    return exports.OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT + contentTokenCount;
 }
 exports.countOpenAIChatMessageTokens = countOpenAIChatMessageTokens;
 async function countOpenAIChatPromptTokens({ messages, model, }) {

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

@@ -1,5 +1,6 @@
 import { countTokens } from "../../../model-function/tokenize-text/countTokens.js";
 import { TikTokenTokenizer } from "../TikTokenTokenizer.js";
+import { getOpenAIChatModelInformation, } from "./OpenAIChatModel.js";
 /**
  * Prompt tokens that are included automatically for every full
  * chat prompt (several messages) that is sent to OpenAI.
@@ -11,8 +12,10 @@ export const OPENAI_CHAT_PROMPT_BASE_TOKEN_COUNT = 2;
  */
 export const OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT = 5;
 export async function countOpenAIChatMessageTokens({ message, model, }) {
-    return (OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT +
-        (await countTokens(new TikTokenTokenizer({ model }), message.content ?? "")));
+    const contentTokenCount = await countTokens(new TikTokenTokenizer({
+        model: getOpenAIChatModelInformation(model).baseModel,
+    }), message.content ?? "");
+    return OPENAI_CHAT_MESSAGE_BASE_TOKEN_COUNT + contentTokenCount;
 }
 export async function countOpenAIChatPromptTokens({ messages, model, }) {
     let tokens = OPENAI_CHAT_PROMPT_BASE_TOKEN_COUNT;

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "Build AI applications, chatbots, and agents with JavaScript and TypeScript.",
-  "version": "0.23.0",
+  "version": "0.24.1",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [