modelfusion 0.98.0 → 0.99.0

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

@@ -32,19 +32,6 @@ export interface AbstractOpenAIChatCallSettings {
             name: string;
         };
     };
-    /**
-     * An array of strings or a single string that the model will recognize as end-of-text indicators.
-     * The model stops generating more content when it encounters any of these strings.
-     * This is particularly useful in scripted or formatted text generation, where a specific end point is required.
-     * Example: stop: ['\n', 'END']
-     */
-    stop?: string | string[];
-    /**
-     * Specifies the maximum number of tokens (words, punctuation, parts of words) that the model can generate in a single response.
-     * It helps to control the length of the output, this can help prevent wasted time and tokens when tweaker topP or temperature.
-     * Example: maxTokens: 1000
-     */
-    maxTokens?: number;
     /**
      * `temperature`: Controls the randomness and creativity in the model's responses.
      * A lower temperature (close to 0) results in more predictable, conservative text, while a higher temperature (close to 1) produces more varied and creative output.
@@ -67,17 +54,6 @@ export interface AbstractOpenAIChatCallSettings {
      * Example: seed: 89 (or) seed: null
      */
     seed?: number | null;
-    responseFormat?: {
-        type?: "text" | "json_object";
-    };
-    /**
-     * Specifies the number of responses or completions the model should generate for a given prompt.
-     * This is useful when you need multiple different outputs or ideas for a single prompt.
-     * The model will generate 'n' distinct responses, each based on the same initial prompt.
-     * In a streaming model this will result in both responses streamed back in real time.
-     * Example: n: 3 // The model will produce 3 different responses.
-     */
-    n?: number;
     /**
      * Discourages the model from repeating the same information or context already mentioned in the conversation or prompt.
      * Increasing this value encourages the model to introduce new topics or ideas, rather than reiterating what has been said.
@@ -92,9 +68,12 @@ export interface AbstractOpenAIChatCallSettings {
      * Example: frequencyPenalty: 0.5 // Moderately discourages repetitive language.
      */
     frequencyPenalty?: number;
+    responseFormat?: {
+        type?: "text" | "json_object";
+    };
     logitBias?: Record<number, number>;
 }
-export interface AbstractOpenAIChatSettings extends TextGenerationModelSettings, Omit<AbstractOpenAIChatCallSettings, "stop" | "maxTokens"> {
+export interface AbstractOpenAIChatSettings extends TextGenerationModelSettings, AbstractOpenAIChatCallSettings {
     isUserIdForwardingEnabled?: boolean;
 }
 export type OpenAIChatPrompt = OpenAIChatMessage[];
@@ -113,7 +92,7 @@ export declare abstract class AbstractOpenAIChatModel<SETTINGS extends AbstractO
         tools?: AbstractOpenAIChatCallSettings["tools"];
         toolChoice?: AbstractOpenAIChatCallSettings["toolChoice"];
     }): Promise<RESULT>;
-    doGenerateText(prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<{
+    doGenerateTexts(prompt: OpenAIChatPrompt, options?: FunctionOptions): Promise<{
         response: {
             object: "chat.completion";
             usage: {
@@ -147,7 +126,7 @@ export declare abstract class AbstractOpenAIChatModel<SETTINGS extends AbstractO
             }[];
             system_fingerprint?: string | null | undefined;
         };
-        text: string;
+        texts: string[];
         usage: {
             promptTokens: number;
             completionTokens: number;

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

@@ -17,38 +17,67 @@ export class AbstractOpenAIChatModel extends AbstractModel {
         super({ settings });
     }
     async callAPI(messages, options) {
+        const api = this.settings.api ?? new OpenAIApiConfiguration();
+        const responseFormat = options.responseFormat;
+        const abortSignal = options.run?.abortSignal;
+        const user = this.settings.isUserIdForwardingEnabled
+            ? options.run?.userId
+            : undefined;
+        const openAIResponseFormat = this.settings.responseFormat;
+        // function & tool calling:
+        const functions = options.functions ?? this.settings.functions;
+        const functionCall = options.functionCall ?? this.settings.functionCall;
+        const tools = options.tools ?? this.settings.tools;
+        const toolChoice = options.toolChoice ?? this.settings.toolChoice;
+        let { stopSequences } = this.settings;
         return callWithRetryAndThrottle({
             retry: this.settings.api?.retry,
             throttle: this.settings.api?.throttle,
-            call: async () => callOpenAIChatCompletionAPI({
-                ...this.settings,
-                // function & tool calling:
-                functions: options.functions ?? this.settings.functions,
-                functionCall: options.functionCall ?? this.settings.functionCall,
-                tools: options.tools ?? this.settings.tools,
-                toolChoice: options.toolChoice ?? this.settings.toolChoice,
-                // map to OpenAI API names:
-                stop: this.settings.stopSequences,
-                maxTokens: this.settings.maxCompletionTokens,
-                openAIResponseFormat: this.settings.responseFormat,
-                // other settings:
-                user: this.settings.isUserIdForwardingEnabled
-                    ? options.run?.userId
-                    : undefined,
-                abortSignal: options.run?.abortSignal,
-                responseFormat: options.responseFormat,
-                messages,
-            }),
+            call: async () => {
+                // empty arrays are not allowed for stopSequences:
+                if (stopSequences != null &&
+                    Array.isArray(stopSequences) &&
+                    stopSequences.length === 0) {
+                    stopSequences = undefined;
+                }
+                return postJsonToApi({
+                    url: api.assembleUrl("/chat/completions"),
+                    headers: api.headers,
+                    body: {
+                        stream: responseFormat.stream,
+                        model: this.settings.model,
+                        messages,
+                        functions,
+                        function_call: functionCall,
+                        tools,
+                        tool_choice: toolChoice,
+                        temperature: this.settings.temperature,
+                        top_p: this.settings.topP,
+                        n: this.settings.numberOfGenerations,
+                        stop: this.settings.stopSequences,
+                        max_tokens: this.settings.maxGenerationTokens,
+                        presence_penalty: this.settings.presencePenalty,
+                        frequency_penalty: this.settings.frequencyPenalty,
+                        logit_bias: this.settings.logitBias,
+                        seed: this.settings.seed,
+                        response_format: openAIResponseFormat,
+                        user,
+                    },
+                    failedResponseHandler: failedOpenAICallResponseHandler,
+                    successfulResponseHandler: responseFormat.handler,
+                    abortSignal,
+                });
+            },
         });
     }
-    async doGenerateText(prompt, options) {
+    async doGenerateTexts(prompt, options) {
         const response = await this.callAPI(prompt, {
             ...options,
             responseFormat: OpenAIChatResponseFormat.json,
         });
         return {
             response,
-            text: response.choices[0].message.content,
+            texts: response.choices.map((choice) => choice.message.content ?? ""),
             usage: this.extractUsage(response),
         };
     }
@@ -169,39 +198,6 @@ const openAIChatResponseSchema = z.object({
         total_tokens: z.number(),
     }),
 });
-async function callOpenAIChatCompletionAPI({ api = new OpenAIApiConfiguration(), abortSignal, responseFormat, model, messages, functions, functionCall, tools, toolChoice, temperature, topP, n, stop, maxTokens, presencePenalty, frequencyPenalty, logitBias, user, openAIResponseFormat, seed, }) {
-    // empty arrays are not allowed for stop:
-    if (stop != null && Array.isArray(stop) && stop.length === 0) {
-        stop = undefined;
-    }
-    return postJsonToApi({
-        url: api.assembleUrl("/chat/completions"),
-        headers: api.headers,
-        body: {
-            stream: responseFormat.stream,
-            model,
-            messages,
-            functions,
-            function_call: functionCall,
-            tools,
-            tool_choice: toolChoice,
-            temperature,
-            top_p: topP,
-            n,
-            stop,
-            max_tokens: maxTokens,
-            presence_penalty: presencePenalty,
-            frequency_penalty: frequencyPenalty,
-            logit_bias: logitBias,
-            seed,
-            response_format: openAIResponseFormat,
-            user,
-        },
-        failedResponseHandler: failedOpenAICallResponseHandler,
-        successfulResponseHandler: responseFormat.handler,
-        abortSignal,
-    });
-}
 export const OpenAIChatResponseFormat = {
     /**
      * Returns the response as a JSON object.

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

@@ -144,7 +144,7 @@ exports.calculateOpenAIChatCostInMillicents = calculateOpenAIChatCostInMillicent
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -196,13 +196,13 @@ class OpenAIChatModel extends AbstractOpenAIChatModel_js_1.AbstractOpenAIChatMod
     }
     get settingsForEvent() {
         const eventSettingProperties = [
+            "maxGenerationTokens",
             "stopSequences",
-            "maxCompletionTokens",
+            "numberOfGenerations",
             "functions",
             "functionCall",
             "temperature",
             "topP",
-            "n",
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",

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

@@ -117,7 +117,7 @@ export interface OpenAIChatSettings extends TextGenerationModelSettings, Omit<Op
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([

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

@@ -138,7 +138,7 @@ export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => {
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -190,13 +190,13 @@ export class OpenAIChatModel extends AbstractOpenAIChatModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
+            "maxGenerationTokens",
             "stopSequences",
-            "maxCompletionTokens",
+            "numberOfGenerations",
             "functions",
             "functionCall",
             "temperature",
             "topP",
-            "n",
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",

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

@@ -50,7 +50,7 @@ describe("streamText", () => {
         const stream = await (0, streamText_js_1.streamText)(new OpenAIChatModel_js_1.OpenAIChatModel({
             api: new OpenAIApiConfiguration_js_1.OpenAIApiConfiguration({ apiKey: "test" }),
             model: "gpt-3.5-turbo",
-            n: 2,
+            numberOfGenerations: 2,
         }).withTextPrompt(), "test prompt");
         const chunks = [];
         for await (const part of stream) {

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

@@ -48,7 +48,7 @@ describe("streamText", () => {
         const stream = await streamText(new OpenAIChatModel({
             api: new OpenAIApiConfiguration({ apiKey: "test" }),
             model: "gpt-3.5-turbo",
-            n: 2,
+            numberOfGenerations: 2,
         }).withTextPrompt(), "test prompt");
         const chunks = [];
         for await (const part of stream) {

package/model-provider/openai-compatible/OpenAICompatibleChatModel.cjs

@@ -44,12 +44,12 @@ class OpenAICompatibleChatModel extends AbstractOpenAIChatModel_js_1.AbstractOpe
     get settingsForEvent() {
         const eventSettingProperties = [
             "stopSequences",
-            "maxCompletionTokens",
+            "maxGenerationTokens",
+            "numberOfGenerations",
             "functions",
             "functionCall",
             "temperature",
             "topP",
-            "n",
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",

package/model-provider/openai-compatible/OpenAICompatibleChatModel.js

@@ -41,12 +41,12 @@ export class OpenAICompatibleChatModel extends AbstractOpenAIChatModel {
     get settingsForEvent() {
         const eventSettingProperties = [
             "stopSequences",
-            "maxCompletionTokens",
+            "maxGenerationTokens",
+            "numberOfGenerations",
             "functions",
             "functionCall",
             "temperature",
             "topP",
-            "n",
             "presencePenalty",
             "frequencyPenalty",
             "logitBias",

package/model-provider/openai-compatible/OpenAICompatibleFacade.cjs

@@ -14,7 +14,7 @@ const OpenAICompatibleChatModel_js_1 = require("./OpenAICompatibleChatModel.cjs"
  * const model = openaicompatible.ChatTextGenerator({
  *   model: "provider-specific-model-name",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([

package/model-provider/openai-compatible/OpenAICompatibleFacade.d.ts

@@ -11,7 +11,7 @@ import { OpenAICompatibleChatModel, OpenAICompatibleChatSettings } from "./OpenA
  * const model = openaicompatible.ChatTextGenerator({
  *   model: "provider-specific-model-name",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([

package/model-provider/openai-compatible/OpenAICompatibleFacade.js

@@ -11,7 +11,7 @@ import { OpenAICompatibleChatModel, } from "./OpenAICompatibleChatModel.js";
  * const model = openaicompatible.ChatTextGenerator({
  *   model: "provider-specific-model-name",
  *   temperature: 0.7,
- *   maxCompletionTokens: 500,
+ *   maxGenerationTokens: 500,
  * });
  *
  * const text = await generateText([

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "The TypeScript library for building multi-modal AI applications.",
-  "version": "0.98.0",
+  "version": "0.99.0",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [