modelfusion 0.14.0 → 0.16.0

package/README.md

@@ -58,7 +58,10 @@ const text = await generateText(
 
 ```ts
 const textStream = await streamText(
-  new OpenAIChatModel({ model: "gpt-3.5-turbo", maxTokens: 1000 }),
+  new OpenAIChatModel({
+    model: "gpt-3.5-turbo",
+    maxCompletionTokens: 1000,
+  }),
   [
     OpenAIChatMessage.system("You are a story writer."),
     OpenAIChatMessage.user("Write a story about a robot learning to love"),
@@ -78,7 +81,7 @@ for await (const textFragment of textStream) {
 const text = await generateText(
   new LlamaCppTextGenerationModel({
     contextWindowSize: 4096, // Llama 2 context window size
-    nPredict: 1000,
+    maxCompletionTokens: 1000,
   }).withPromptFormat(Llama2InstructionPromptFormat()),
   {
     system: "You are a story writer.",
@@ -111,7 +114,7 @@ ModelFusion model functions return rich results that include the original respon
 const { response, metadata } = await generateText(
   new OpenAITextGenerationModel({
     model: "text-davinci-003",
-    maxTokens: 1000,
+    maxCompletionTokens: 1000,
     n: 2, // generate 2 completions
   }),
   "Write a short story about a robot learning to love:\n\n",
@@ -134,7 +137,7 @@ const value = await generateJson(
   new OpenAIChatModel({
     model: "gpt-3.5-turbo",
     temperature: 0,
-    maxTokens: 50,
+    maxCompletionTokens: 50,
   }),
   {
     name: "sentiment" as const,
@@ -165,7 +168,10 @@ It either matches one of the schemas or is text reponse.
 
 ```ts
 const { schema, value, text } = await generateJsonOrText(
-  new OpenAIChatModel({ model: "gpt-3.5-turbo", maxTokens: 1000 }),
+  new OpenAIChatModel({
+    model: "gpt-3.5-turbo",
+    maxCompletionTokens: 1000,
+  }),
   [
     {
       name: "getCurrentWeather" as const, // mark 'as const' for type inference

package/composed-function/summarize/summarizeRecursivelyWithTextGenerationAndTokenSplitting.cjs

@@ -10,7 +10,7 @@ const summarizeRecursively_js_1 = require("./summarizeRecursively.cjs");
  * while leaving enough space for the model to generate text.
  */
 async function summarizeRecursivelyWithTextGenerationAndTokenSplitting({ text, model, prompt, tokenLimit = model.contextWindowSize -
-    (model.maxCompletionTokens ?? model.contextWindowSize / 4), join, }, options) {
+    (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), join, }, options) {
     const emptyPromptTokens = await model.countPromptTokens(await prompt({ text: "" }));
     return (0, summarizeRecursively_js_1.summarizeRecursively)({
         split: (0, splitRecursively_js_1.splitAtToken)({

package/composed-function/summarize/summarizeRecursivelyWithTextGenerationAndTokenSplitting.js

@@ -7,7 +7,7 @@ import { summarizeRecursively } from "./summarizeRecursively.js";
  * while leaving enough space for the model to generate text.
  */
 export async function summarizeRecursivelyWithTextGenerationAndTokenSplitting({ text, model, prompt, tokenLimit = model.contextWindowSize -
-    (model.maxCompletionTokens ?? model.contextWindowSize / 4), join, }, options) {
+    (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), join, }, options) {
     const emptyPromptTokens = await model.countPromptTokens(await prompt({ text: "" }));
     return summarizeRecursively({
         split: splitAtToken({

package/model-function/Model.d.ts

@@ -12,11 +12,11 @@ export interface Model<SETTINGS> {
      * @example
      * const model = new OpenAITextGenerationModel({
      *   model: "text-davinci-003",
-     *   maxTokens: 500,
+     *   maxCompletionTokens: 500,
      * });
      *
      * const modelWithMoreTokens = model.withSettings({
-     *   maxTokens: 1000,
+     *   maxCompletionTokens: 1000,
      * });
      */
     withSettings(additionalSettings: Partial<SETTINGS>): this;

package/model-function/generate-text/TextGenerationModel.d.ts

@@ -5,7 +5,21 @@ import { Model, ModelSettings } from "../Model.js";
 import { BasicTokenizer, FullTokenizer } from "../tokenize-text/Tokenizer.js";
 import { DeltaEvent } from "./DeltaEvent.js";
 export interface TextGenerationModelSettings extends ModelSettings {
-    trimOutput?: boolean;
+    /**
+     * Maximum number of tokens to generate.
+     * Does nothing if the model does not support this setting.
+     */
+    maxCompletionTokens?: number | undefined;
+    /**
+     * Stop sequences to use. Stop sequences are not included in the generated text.
+     * Does nothing if the model does not support this setting.
+     */
+    stopSequences?: string[] | undefined;
+    /**
+     * When true, the leading and trailing white space and line terminator characters
+     * are removed from the generated text.
+     */
+    trimWhitespace?: boolean;
 }
 export interface TextGenerationModel<PROMPT, RESPONSE, FULL_DELTA, SETTINGS extends TextGenerationModelSettings> extends Model<SETTINGS> {
     readonly contextWindowSize: number | undefined;
@@ -25,18 +39,4 @@ export interface TextGenerationModel<PROMPT, RESPONSE, FULL_DELTA, SETTINGS exte
      */
     readonly extractTextDelta: ((fullDelta: FULL_DELTA) => string | undefined) | undefined;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, PROMPT>): PromptFormatTextGenerationModel<INPUT_PROMPT, PROMPT, RESPONSE, FULL_DELTA, SETTINGS, this>;
-    /**
-     * Maximum number of tokens to generate.
-     */
-    readonly maxCompletionTokens: number | undefined;
-    /**
-     * Sets the maximum number of tokens to generate.
-     * Does nothing if the model does not support this setting.
-     */
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    /**
-     * Sets the stop tokens to use. Stop tokens are not included in the generated text.
-     * Does nothing if the model does not support this setting.
-     */
-    withStopTokens(stopTokens: string[]): this;
 }

package/model-function/generate-text/generateText.cjs

@@ -10,8 +10,8 @@ model, prompt, options) {
         options,
         generateResponse: (options) => model.generateTextResponse(prompt, options),
         extractOutputValue: (result) => {
-            const shouldTrimOutput = model.settings.trimOutput ?? true;
-            return shouldTrimOutput
+            const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
+            return shouldTrimWhitespace
                 ? model.extractText(result).trim()
                 : model.extractText(result);
         },

package/model-function/generate-text/generateText.js

@@ -7,8 +7,8 @@ model, prompt, options) {
         options,
         generateResponse: (options) => model.generateTextResponse(prompt, options),
         extractOutputValue: (result) => {
-            const shouldTrimOutput = model.settings.trimOutput ?? true;
-            return shouldTrimOutput
+            const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
+            return shouldTrimWhitespace
                 ? model.extractText(result).trim()
                 : model.extractText(result);
         },

package/model-provider/cohere/CohereTextGenerationModel.cjs

@@ -37,7 +37,7 @@ exports.COHERE_TEXT_GENERATION_MODELS = {
  * const model = new CohereTextGenerationModel({
  *   model: "command-nightly",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText(
@@ -91,13 +91,21 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            // use endSequences instead of stopSequences
+            // to exclude stop tokens from the generated text
+            endSequences: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -124,24 +132,15 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new CohereTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        // use endSequences instead of stopSequences
-        // to exclude stop tokens from the generated text
-        return this.withSettings({ endSequences: stopTokens });
-    }
 }
 exports.CohereTextGenerationModel = CohereTextGenerationModel;
 const cohereTextGenerationResponseSchema = zod_1.z.object({

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

@@ -35,13 +35,11 @@ export interface CohereTextGenerationModelSettings extends TextGenerationModelSe
         throttle?: ThrottleFunction;
     };
     numGenerations?: number;
-    maxTokens?: number;
     temperature?: number;
     k?: number;
     p?: number;
     frequencyPenalty?: number;
     presencePenalty?: number;
-    endSequences?: string[];
     stopSequences?: string[];
     returnLikelihoods?: "GENERATION" | "ALL" | "NONE";
     logitBias?: Record<string, number>;
@@ -56,7 +54,7 @@ export interface CohereTextGenerationModelSettings extends TextGenerationModelSe
  * const model = new CohereTextGenerationModel({
  *   model: "command-nightly",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText(
@@ -94,9 +92,6 @@ export declare class CohereTextGenerationModel extends AbstractModel<CohereTextG
     extractTextDelta(fullDelta: CohereTextGenerationDelta): string | undefined;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, CohereTextGenerationResponse, CohereTextGenerationDelta, CohereTextGenerationModelSettings, this>;
     withSettings(additionalSettings: Partial<CohereTextGenerationModelSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const cohereTextGenerationResponseSchema: z.ZodObject<{
     id: z.ZodString;

package/model-provider/cohere/CohereTextGenerationModel.js

@@ -31,7 +31,7 @@ export const COHERE_TEXT_GENERATION_MODELS = {
  * const model = new CohereTextGenerationModel({
  *   model: "command-nightly",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText(
@@ -85,13 +85,21 @@ export class CohereTextGenerationModel extends AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            // use endSequences instead of stopSequences
+            // to exclude stop tokens from the generated text
+            endSequences: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -118,24 +126,15 @@ export class CohereTextGenerationModel extends AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new CohereTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        // use endSequences instead of stopSequences
-        // to exclude stop tokens from the generated text
-        return this.withSettings({ endSequences: stopTokens });
-    }
 }
 const cohereTextGenerationResponseSchema = z.object({
     id: z.string(),

package/model-provider/huggingface/HuggingFaceTextGenerationModel.cjs

@@ -19,7 +19,7 @@ const PromptFormatTextGenerationModel_js_1 = require("../../prompt/PromptFormatT
  * const model = new HuggingFaceTextGenerationModel({
  *   model: "tiiuae/falcon-7b",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -81,16 +81,21 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     async callAPI(prompt, options) {
         const run = options?.run;
         const settings = options?.settings;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             options: {
                 useCache: true,
                 waitForModel: true,
             },
-        }, this.settings, settings, {
+            ...combinedSettings,
+            maxNewTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             inputs: prompt,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -112,16 +117,6 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     withSettings(additionalSettings) {
         return new HuggingFaceTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxNewTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxNewTokens: maxCompletionTokens });
-    }
-    withStopTokens() {
-        // stop tokens are not supported by the HuggingFace API
-        return this;
-    }
 }
 exports.HuggingFaceTextGenerationModel = HuggingFaceTextGenerationModel;
 const huggingFaceTextGenerationResponseSchema = zod_1.default.array(zod_1.default.object({

package/model-provider/huggingface/HuggingFaceTextGenerationModel.d.ts

@@ -16,7 +16,6 @@ export interface HuggingFaceTextGenerationModelSettings extends TextGenerationMo
     topP?: number;
     temperature?: number;
     repetitionPenalty?: number;
-    maxNewTokens?: number;
     maxTime?: number;
     numReturnSequences?: number;
     doSample?: boolean;
@@ -34,7 +33,7 @@ export interface HuggingFaceTextGenerationModelSettings extends TextGenerationMo
  * const model = new HuggingFaceTextGenerationModel({
  *   model: "tiiuae/falcon-7b",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -60,9 +59,6 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
     extractTextDelta: undefined;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, HuggingFaceTextGenerationResponse, undefined, HuggingFaceTextGenerationModelSettings, this>;
     withSettings(additionalSettings: Partial<HuggingFaceTextGenerationModelSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(): this;
 }
 declare const huggingFaceTextGenerationResponseSchema: z.ZodArray<z.ZodObject<{
     generated_text: z.ZodString;

package/model-provider/huggingface/HuggingFaceTextGenerationModel.js

@@ -13,7 +13,7 @@ import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGe
  * const model = new HuggingFaceTextGenerationModel({
  *   model: "tiiuae/falcon-7b",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -75,16 +75,21 @@ export class HuggingFaceTextGenerationModel extends AbstractModel {
     async callAPI(prompt, options) {
         const run = options?.run;
         const settings = options?.settings;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             options: {
                 useCache: true,
                 waitForModel: true,
             },
-        }, this.settings, settings, {
+            ...combinedSettings,
+            maxNewTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             inputs: prompt,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -106,16 +111,6 @@ export class HuggingFaceTextGenerationModel extends AbstractModel {
     withSettings(additionalSettings) {
         return new HuggingFaceTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxNewTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxNewTokens: maxCompletionTokens });
-    }
-    withStopTokens() {
-        // stop tokens are not supported by the HuggingFace API
-        return this;
-    }
 }
 const huggingFaceTextGenerationResponseSchema = z.array(z.object({
     generated_text: z.string(),

package/model-provider/llamacpp/LlamaCppTextGenerationModel.cjs

@@ -43,11 +43,18 @@ class LlamaCppTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign(this.settings, settings, {
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
+            ...combinedSettings,
+            nPredict: combinedSettings.maxCompletionTokens,
+            stop: combinedSettings.stopSequences,
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -78,22 +85,15 @@ class LlamaCppTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new LlamaCppTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.nPredict;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ nPredict: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 exports.LlamaCppTextGenerationModel = LlamaCppTextGenerationModel;
 const llamaCppTextGenerationResponseSchema = zod_1.default.object({

package/model-provider/llamacpp/LlamaCppTextGenerationModel.d.ts

@@ -25,9 +25,7 @@ export interface LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE extends
     temperature?: number;
     topK?: number;
     topP?: number;
-    nPredict?: number;
     nKeep?: number;
-    stop?: string[];
     tfsZ?: number;
     typicalP?: number;
     repeatPenalty?: number;
@@ -59,8 +57,8 @@ export declare class LlamaCppTextGenerationModel<CONTEXT_WINDOW_SIZE extends num
             model: string;
             stream: boolean;
             seed: number;
-            stop: string[];
             mirostat: number;
+            stop: string[];
             frequency_penalty: number;
             ignore_eos: boolean;
             logit_bias: number[];
@@ -104,9 +102,6 @@ export declare class LlamaCppTextGenerationModel<CONTEXT_WINDOW_SIZE extends num
     extractTextDelta(fullDelta: LlamaCppTextGenerationDelta): string | undefined;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, LlamaCppTextGenerationResponse, LlamaCppTextGenerationDelta, LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>, this>;
     withSettings(additionalSettings: Partial<LlamaCppTextGenerationModelSettings<CONTEXT_WINDOW_SIZE>>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
     content: z.ZodString;
@@ -139,8 +134,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -163,8 +158,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -231,8 +226,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -279,8 +274,8 @@ declare const llamaCppTextGenerationResponseSchema: z.ZodObject<{
         model: string;
         stream: boolean;
         seed: number;
-        stop: string[];
         mirostat: number;
+        stop: string[];
         frequency_penalty: number;
         ignore_eos: boolean;
         logit_bias: number[];
@@ -344,8 +339,8 @@ export declare const LlamaCppTextGenerationResponseFormat: {
                 model: string;
                 stream: boolean;
                 seed: number;
-                stop: string[];
                 mirostat: number;
+                stop: string[];
                 frequency_penalty: number;
                 ignore_eos: boolean;
                 logit_bias: number[];

package/model-provider/llamacpp/LlamaCppTextGenerationModel.js

@@ -37,11 +37,18 @@ export class LlamaCppTextGenerationModel extends AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign(this.settings, settings, {
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
+            ...combinedSettings,
+            nPredict: combinedSettings.maxCompletionTokens,
+            stop: combinedSettings.stopSequences,
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: this.settings.retry,
             throttle: this.settings.throttle,
@@ -72,22 +79,15 @@ export class LlamaCppTextGenerationModel extends AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new LlamaCppTextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.nPredict;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ nPredict: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const llamaCppTextGenerationResponseSchema = z.object({
     content: z.string(),

package/model-provider/openai/OpenAITextGenerationModel.cjs

@@ -75,7 +75,7 @@ exports.calculateOpenAITextGenerationCostInMillicents = calculateOpenAITextGener
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -124,14 +124,23 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            // Copied settings:
+            ...combinedSettings,
+            // map to OpenAI API names:
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
+            // other settings:
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -158,22 +167,15 @@ class OpenAITextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAITextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 exports.OpenAITextGenerationModel = OpenAITextGenerationModel;
 const openAITextGenerationResponseSchema = zod_1.default.object({

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

@@ -72,13 +72,11 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
     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;
@@ -92,7 +90,7 @@ export interface OpenAITextGenerationModelSettings extends TextGenerationModelSe
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -136,9 +134,6 @@ export declare class OpenAITextGenerationModel extends AbstractModel<OpenAITextG
     extractTextDelta(fullDelta: OpenAITextGenerationDelta): string | undefined;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, string>): PromptFormatTextGenerationModel<INPUT_PROMPT, string, OpenAITextGenerationResponse, OpenAITextGenerationDelta, OpenAITextGenerationModelSettings, this>;
     withSettings(additionalSettings: Partial<OpenAITextGenerationModelSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const openAITextGenerationResponseSchema: z.ZodObject<{
     id: z.ZodString;

package/model-provider/openai/OpenAITextGenerationModel.js

@@ -67,7 +67,7 @@ export const calculateOpenAITextGenerationCostInMillicents = ({ model, response,
  * const model = new OpenAITextGenerationModel({
  *   model: "text-davinci-003",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  *   retry: retryWithExponentialBackoff({ maxTries: 5 }),
  * });
  *
@@ -116,14 +116,23 @@ export class OpenAITextGenerationModel extends AbstractModel {
     }
     async callAPI(prompt, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            // Copied settings:
+            ...combinedSettings,
+            // map to OpenAI API names:
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
+            // other settings:
             abortSignal: run?.abortSignal,
             prompt,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -150,22 +159,15 @@ export class OpenAITextGenerationModel extends AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAITextGenerationModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const openAITextGenerationResponseSchema = z.object({
     id: z.string(),

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

@@ -93,7 +93,7 @@ exports.calculateOpenAIChatCostInMillicents = calculateOpenAIChatCostInMillicent
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -150,14 +150,20 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
     }
     async callAPI(messages, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             messages,
             responseFormat,
-        });
+        };
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -207,22 +213,13 @@ class OpenAIChatModel extends AbstractModel_js_1.AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({ stopSequences: promptFormat.stopSequences }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAIChatModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 exports.OpenAIChatModel = OpenAIChatModel;
 const openAIChatResponseSchema = zod_1.default.object({

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

@@ -95,7 +95,7 @@ export interface OpenAIChatCallSettings {
     presencePenalty?: number;
     frequencyPenalty?: number;
 }
-export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIModelSettings, OpenAIChatCallSettings {
+export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIModelSettings, Omit<OpenAIChatCallSettings, "stop" | "maxTokens"> {
     isUserIdForwardingEnabled?: boolean;
 }
 /**
@@ -107,7 +107,7 @@ export interface OpenAIChatSettings extends TextGenerationModelSettings, OpenAIM
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -172,9 +172,6 @@ export declare class OpenAIChatModel extends AbstractModel<OpenAIChatSettings> i
     extractJson(response: OpenAIChatResponse): unknown;
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, OpenAIChatMessage[]>): PromptFormatTextGenerationModel<INPUT_PROMPT, OpenAIChatMessage[], OpenAIChatResponse, OpenAIChatDelta, OpenAIChatSettings, this>;
     withSettings(additionalSettings: Partial<OpenAIChatSettings>): this;
-    get maxCompletionTokens(): number | undefined;
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }
 declare const openAIChatResponseSchema: z.ZodObject<{
     id: z.ZodString;

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

@@ -85,7 +85,7 @@ export const calculateOpenAIChatCostInMillicents = ({ model, response, }) => res
  * const model = new OpenAIChatModel({
  *   model: "gpt-3.5-turbo",
  *   temperature: 0.7,
- *   maxTokens: 500,
+ *   maxCompletionTokens: 500,
  * });
  *
  * const text = await generateText([
@@ -142,14 +142,20 @@ export class OpenAIChatModel extends AbstractModel {
     }
     async callAPI(messages, options) {
         const { run, settings, responseFormat } = options;
-        const callSettings = Object.assign({
+        const combinedSettings = {
+            ...this.settings,
+            ...settings,
+        };
+        const callSettings = {
             apiKey: this.apiKey,
             user: this.settings.isUserIdForwardingEnabled ? run?.userId : undefined,
-        }, this.settings, settings, {
+            ...combinedSettings,
+            stop: combinedSettings.stopSequences,
+            maxTokens: combinedSettings.maxCompletionTokens,
             abortSignal: run?.abortSignal,
             messages,
             responseFormat,
-        });
+        };
         return callWithRetryAndThrottle({
             retry: callSettings.retry,
             throttle: callSettings.throttle,
@@ -199,22 +205,13 @@ export class OpenAIChatModel extends AbstractModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({ stopSequences: promptFormat.stopSequences }),
             promptFormat,
         });
     }
     withSettings(additionalSettings) {
         return new OpenAIChatModel(Object.assign({}, this.settings, additionalSettings));
     }
-    get maxCompletionTokens() {
-        return this.settings.maxTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return this.withSettings({ maxTokens: maxCompletionTokens });
-    }
-    withStopTokens(stopTokens) {
-        return this.withSettings({ stop: stopTokens });
-    }
 }
 const openAIChatResponseSchema = z.object({
     id: z.string(),

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "Build AI applications, chatbots, and agents with JavaScript and TypeScript.",
-  "version": "0.14.0",
+  "version": "0.16.0",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [
@@ -67,11 +67,11 @@
     "eslint": "^8.45.0",
     "eslint-config-prettier": "9.0.0",
     "husky": "^8.0.3",
-    "lint-staged": "14.0.0",
+    "lint-staged": "14.0.1",
     "prettier": "3.0.2",
     "rimraf": "5.0.1",
     "typescript": "5.1.6",
-    "zod": "3.22.1",
+    "zod": "3.22.2",
     "zod-to-json-schema": "3.21.4"
   },
   "peerDependencies": {

package/prompt/AlpacaPromptFormat.cjs

@@ -12,7 +12,7 @@ const DEFAULT_SYSTEM_PROMPT_NO_INPUT = "Below is an instruction that describes a
  * @see https://github.com/tatsu-lab/stanford_alpaca#data-release
  */
 const AlpacaInstructionPromptFormat = () => ({
-    stopTokens: [],
+    stopSequences: [],
     format: (instruction) => {
         let text = instruction.system ??
             (instruction.input != null

package/prompt/AlpacaPromptFormat.js

@@ -9,7 +9,7 @@ const DEFAULT_SYSTEM_PROMPT_NO_INPUT = "Below is an instruction that describes a
  * @see https://github.com/tatsu-lab/stanford_alpaca#data-release
  */
 export const AlpacaInstructionPromptFormat = () => ({
-    stopTokens: [],
+    stopSequences: [],
     format: (instruction) => {
         let text = instruction.system ??
             (instruction.input != null

package/prompt/Llama2PromptFormat.cjs

@@ -15,7 +15,7 @@ const END_SYSTEM = "\n<</SYS>>\n\n";
  * @see https://www.philschmid.de/llama-2#how-to-prompt-llama-2-chat
  */
 const Llama2InstructionPromptFormat = () => ({
-    stopTokens: [END_SEGMENT],
+    stopSequences: [END_SEGMENT],
     format: (instruction) => `${BEGIN_SEGMENT}${BEGIN_INSTRUCTION}${instruction.system != null
         ? ` ${BEGIN_SYSTEM}${instruction.system}${END_SYSTEM}`
         : ""} ${instruction.instruction}${instruction.input != null ? `\n\n${instruction.input}` : ""} ${END_INSTRUCTION}\n`,
@@ -54,6 +54,6 @@ const Llama2ChatPromptFormat = () => ({
         }
         return text;
     },
-    stopTokens: [END_SEGMENT],
+    stopSequences: [END_SEGMENT],
 });
 exports.Llama2ChatPromptFormat = Llama2ChatPromptFormat;

package/prompt/Llama2PromptFormat.js

@@ -12,7 +12,7 @@ const END_SYSTEM = "\n<</SYS>>\n\n";
  * @see https://www.philschmid.de/llama-2#how-to-prompt-llama-2-chat
  */
 export const Llama2InstructionPromptFormat = () => ({
-    stopTokens: [END_SEGMENT],
+    stopSequences: [END_SEGMENT],
     format: (instruction) => `${BEGIN_SEGMENT}${BEGIN_INSTRUCTION}${instruction.system != null
         ? ` ${BEGIN_SYSTEM}${instruction.system}${END_SYSTEM}`
         : ""} ${instruction.instruction}${instruction.input != null ? `\n\n${instruction.input}` : ""} ${END_INSTRUCTION}\n`,
@@ -50,5 +50,5 @@ export const Llama2ChatPromptFormat = () => ({
         }
         return text;
     },
-    stopTokens: [END_SEGMENT],
+    stopSequences: [END_SEGMENT],
 });

package/prompt/OpenAIChatPromptFormat.cjs

@@ -26,7 +26,7 @@ const OpenAIChatInstructionPromptFormat = () => ({
         }
         return messages;
     },
-    stopTokens: [],
+    stopSequences: [],
 });
 exports.OpenAIChatInstructionPromptFormat = OpenAIChatInstructionPromptFormat;
 /**
@@ -69,6 +69,6 @@ const OpenAIChatChatPromptFormat = () => ({
         }
         return messages;
     },
-    stopTokens: [],
+    stopSequences: [],
 });
 exports.OpenAIChatChatPromptFormat = OpenAIChatChatPromptFormat;

package/prompt/OpenAIChatPromptFormat.js

@@ -23,7 +23,7 @@ export const OpenAIChatInstructionPromptFormat = () => ({
         }
         return messages;
     },
-    stopTokens: [],
+    stopSequences: [],
 });
 /**
  * Formats a chat prompt as an OpenAI chat prompt.
@@ -65,5 +65,5 @@ export const OpenAIChatChatPromptFormat = () => ({
         }
         return messages;
     },
-    stopTokens: [],
+    stopSequences: [],
 });

package/prompt/PromptFormat.d.ts

@@ -7,8 +7,8 @@ export interface PromptFormat<SOURCE_PROMPT, TARGET_PROMPT> {
      */
     format(sourcePrompt: SOURCE_PROMPT): TARGET_PROMPT;
     /**
-     * The tokens that should be used as default stop tokens.
+     * The texts that should be used as default stop sequences.
      * This is e.g. important for chat formats.
      */
-    stopTokens: string[];
+    stopSequences: string[];
 }

package/prompt/PromptFormatTextGenerationModel.cjs

@@ -59,7 +59,9 @@ class PromptFormatTextGenerationModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
@@ -69,20 +71,5 @@ class PromptFormatTextGenerationModel {
             promptFormat: this.promptFormat,
         });
     }
-    get maxCompletionTokens() {
-        return this.model.maxCompletionTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return new PromptFormatTextGenerationModel({
-            model: this.model.withMaxCompletionTokens(maxCompletionTokens),
-            promptFormat: this.promptFormat,
-        });
-    }
-    withStopTokens(stopTokens) {
-        return new PromptFormatTextGenerationModel({
-            model: this.model.withStopTokens(stopTokens),
-            promptFormat: this.promptFormat,
-        });
-    }
 }
 exports.PromptFormatTextGenerationModel = PromptFormatTextGenerationModel;

package/prompt/PromptFormatTextGenerationModel.d.ts

@@ -20,7 +20,4 @@ export declare class PromptFormatTextGenerationModel<PROMPT, MODEL_PROMPT, RESPO
     get extractTextDelta(): MODEL["extractTextDelta"];
     withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, PROMPT>): PromptFormatTextGenerationModel<INPUT_PROMPT, PROMPT, RESPONSE, FULL_DELTA, SETTINGS, this>;
     withSettings(additionalSettings: Partial<SETTINGS>): this;
-    get maxCompletionTokens(): MODEL["maxCompletionTokens"];
-    withMaxCompletionTokens(maxCompletionTokens: number): this;
-    withStopTokens(stopTokens: string[]): this;
 }

package/prompt/PromptFormatTextGenerationModel.js

@@ -56,7 +56,9 @@ export class PromptFormatTextGenerationModel {
     }
     withPromptFormat(promptFormat) {
         return new PromptFormatTextGenerationModel({
-            model: this.withStopTokens(promptFormat.stopTokens),
+            model: this.withSettings({
+                stopSequences: promptFormat.stopSequences,
+            }),
             promptFormat,
         });
     }
@@ -66,19 +68,4 @@ export class PromptFormatTextGenerationModel {
             promptFormat: this.promptFormat,
         });
     }
-    get maxCompletionTokens() {
-        return this.model.maxCompletionTokens;
-    }
-    withMaxCompletionTokens(maxCompletionTokens) {
-        return new PromptFormatTextGenerationModel({
-            model: this.model.withMaxCompletionTokens(maxCompletionTokens),
-            promptFormat: this.promptFormat,
-        });
-    }
-    withStopTokens(stopTokens) {
-        return new PromptFormatTextGenerationModel({
-            model: this.model.withStopTokens(stopTokens),
-            promptFormat: this.promptFormat,
-        });
-    }
 }

package/prompt/TextPromptFormat.cjs

@@ -6,7 +6,7 @@ const validateChatPrompt_js_1 = require("./chat/validateChatPrompt.cjs");
  * Formats an instruction prompt as a basic text prompt.
  */
 const TextInstructionPromptFormat = () => ({
-    stopTokens: [],
+    stopSequences: [],
     format: (instruction) => {
         let text = "";
         if (instruction.system != null) {
@@ -56,6 +56,6 @@ const TextChatPromptFormat = ({ user, ai }) => ({
         text += `${ai}:\n`;
         return text;
     },
-    stopTokens: [`\n${user}:`],
+    stopSequences: [`\n${user}:`],
 });
 exports.TextChatPromptFormat = TextChatPromptFormat;

package/prompt/TextPromptFormat.js

@@ -3,7 +3,7 @@ import { validateChatPrompt } from "./chat/validateChatPrompt.js";
  * Formats an instruction prompt as a basic text prompt.
  */
 export const TextInstructionPromptFormat = () => ({
-    stopTokens: [],
+    stopSequences: [],
     format: (instruction) => {
         let text = "";
         if (instruction.system != null) {
@@ -52,5 +52,5 @@ export const TextChatPromptFormat = ({ user, ai }) => ({
         text += `${ai}:\n`;
         return text;
     },
-    stopTokens: [`\n${user}:`],
+    stopSequences: [`\n${user}:`],
 });

package/prompt/VicunaPromptFormat.cjs

@@ -50,6 +50,6 @@ const VicunaChatPromptFormat = () => ({
         text += `ASSISTANT: `;
         return text;
     },
-    stopTokens: [`\nUSER:`],
+    stopSequences: [`\nUSER:`],
 });
 exports.VicunaChatPromptFormat = VicunaChatPromptFormat;

package/prompt/VicunaPromptFormat.js

@@ -47,5 +47,5 @@ export const VicunaChatPromptFormat = () => ({
         text += `ASSISTANT: `;
         return text;
     },
-    stopTokens: [`\nUSER:`],
+    stopSequences: [`\nUSER:`],
 });

package/prompt/chat/trimChatPrompt.cjs

@@ -13,7 +13,7 @@ const validateChatPrompt_js_1 = require("./validateChatPrompt.cjs");
  * @see https://modelfusion.dev/guide/function/generate-text/prompt-format#limiting-the-chat-length
  */
 async function trimChatPrompt({ prompt, model, tokenLimit = model.contextWindowSize -
-    (model.maxCompletionTokens ?? model.contextWindowSize / 4), }) {
+    (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), }) {
     (0, validateChatPrompt_js_1.validateChatPrompt)(prompt);
     const startsWithSystemMessage = "system" in prompt[0];
     const systemMessage = startsWithSystemMessage ? [prompt[0]] : [];

package/prompt/chat/trimChatPrompt.js

@@ -10,7 +10,7 @@ import { validateChatPrompt } from "./validateChatPrompt.js";
  * @see https://modelfusion.dev/guide/function/generate-text/prompt-format#limiting-the-chat-length
  */
 export async function trimChatPrompt({ prompt, model, tokenLimit = model.contextWindowSize -
-    (model.maxCompletionTokens ?? model.contextWindowSize / 4), }) {
+    (model.settings.maxCompletionTokens ?? model.contextWindowSize / 4), }) {
     validateChatPrompt(prompt);
     const startsWithSystemMessage = "system" in prompt[0];
     const systemMessage = startsWithSystemMessage ? [prompt[0]] : [];