modelfusion 0.13.0 → 0.15.0

package/README.md

@@ -43,7 +43,7 @@ You can provide API keys for the different [integrations](https://modelfusion.de
 
 Generate text using a language model and a prompt.
 You can stream the text if it is supported by the model.
-You can use [prompt mappings](https://modelfusion.dev/guide/function/generate-text/prompt-mapping) to change the prompt format of a model.
+You can use [prompt formats](https://modelfusion.dev/guide/function/generate-text/prompt-format) to change the prompt format of a model.
 
 #### generateText
 
@@ -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"),
@@ -70,16 +73,16 @@ for await (const textFragment of textStream) {
 }
 ```
 
-#### Prompt Mapping
+#### Prompt Format
 
-[Prompt mapping](https://modelfusion.dev/guide/function/generate-text/prompt-mapping) lets you use higher level prompt structures (such as instruction or chat prompts) for different models.
+[Prompt format](https://modelfusion.dev/guide/function/generate-text/prompt-format) lets you use higher level prompt structures (such as instruction or chat prompts) for different models.
 
 ```ts
 const text = await generateText(
   new LlamaCppTextGenerationModel({
     contextWindowSize: 4096, // Llama 2 context window size
-    nPredict: 1000,
-  }).mapPrompt(InstructionToLlama2PromptMapping()),
+    maxCompletionTokens: 1000,
+  }).withPromptFormat(Llama2InstructionPromptFormat()),
   {
     system: "You are a story writer.",
     instruction: "Write a short story about a robot learning to love.",
@@ -91,7 +94,7 @@ const text = await generateText(
 const textStream = await streamText(
   new OpenAIChatModel({
     model: "gpt-3.5-turbo",
-  }).mapPrompt(ChatToOpenAIChatPromptMapping()),
+  }).withPromptFormat(OpenAIChatChatPromptFormat()),
   [
     { system: "You are a celebrated poet." },
     { user: "Write a short story about a robot learning to love." },
@@ -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

@@ -1,11 +1,25 @@
-import { PromptMapping } from "../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormat } from "../../prompt/PromptFormat.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { FunctionOptions } from "../FunctionOptions.js";
 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;
@@ -24,19 +38,5 @@ export interface TextGenerationModel<PROMPT, RESPONSE, FULL_DELTA, SETTINGS exte
      * Optional. Implement for streaming support.
      */
     readonly extractTextDelta: ((fullDelta: FULL_DELTA) => string | undefined) | undefined;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, PROMPT>): PromptMappingTextGenerationModel<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;
+    withPromptFormat<INPUT_PROMPT>(promptFormat: PromptFormat<INPUT_PROMPT, PROMPT>): PromptFormatTextGenerationModel<INPUT_PROMPT, PROMPT, RESPONSE, FULL_DELTA, SETTINGS, 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

@@ -9,7 +9,7 @@ const zod_1 = require("zod");
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const AsyncQueue_js_1 = require("../../model-function/generate-text/AsyncQueue.cjs");
 const countTokens_js_1 = require("../../model-function/tokenize-text/countTokens.cjs");
-const PromptMappingTextGenerationModel_js_1 = require("../../prompt/PromptMappingTextGenerationModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../prompt/PromptFormatTextGenerationModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../util/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../util/api/postToApi.cjs");
 const CohereError_js_1 = require("./CohereError.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,
@@ -122,26 +130,17 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta.delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel_js_1.PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
+            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

@@ -3,8 +3,8 @@ import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { FunctionOptions } from "../../model-function/FunctionOptions.js";
 import { DeltaEvent } from "../../model-function/generate-text/DeltaEvent.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../model-function/generate-text/TextGenerationModel.js";
-import { PromptMapping } from "../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.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";
@@ -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(
@@ -92,11 +90,8 @@ export declare class CohereTextGenerationModel extends AbstractModel<CohereTextG
     extractText(response: CohereTextGenerationResponse): string;
     generateDeltaStreamResponse(prompt: string, options?: FunctionOptions<CohereTextGenerationModelSettings>): Promise<AsyncIterable<DeltaEvent<CohereTextGenerationDelta>>>;
     extractTextDelta(fullDelta: CohereTextGenerationDelta): string | undefined;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, string>): PromptMappingTextGenerationModel<INPUT_PROMPT, string, CohereTextGenerationResponse, CohereTextGenerationDelta, CohereTextGenerationModelSettings, this>;
+    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

@@ -3,7 +3,7 @@ import { z } from "zod";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { AsyncQueue } from "../../model-function/generate-text/AsyncQueue.js";
 import { countTokens } from "../../model-function/tokenize-text/countTokens.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 import { callWithRetryAndThrottle } from "../../util/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../util/api/postToApi.js";
 import { failedCohereCallResponseHandler } from "./CohereError.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,
@@ -116,26 +124,17 @@ export class CohereTextGenerationModel extends AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta.delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel({
+            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

@@ -9,7 +9,7 @@ const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../util/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../util/api/postToApi.cjs");
 const HuggingFaceError_js_1 = require("./HuggingFaceError.cjs");
-const PromptMappingTextGenerationModel_js_1 = require("../../prompt/PromptMappingTextGenerationModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../prompt/PromptFormatTextGenerationModel.cjs");
 /**
  * Create a text generation model that calls a Hugging Face Inference API Text Generation Task.
  *
@@ -19,7 +19,7 @@ const PromptMappingTextGenerationModel_js_1 = require("../../prompt/PromptMappin
  * 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,
@@ -103,25 +108,15 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     extractText(response) {
         return response[0].generated_text;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel_js_1.PromptMappingTextGenerationModel({
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
             model: this,
-            promptMapping,
+            promptFormat,
         });
     }
     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

@@ -4,8 +4,8 @@ import { FunctionOptions } from "../../model-function/FunctionOptions.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "../../model-function/generate-text/TextGenerationModel.js";
 import { RetryFunction } from "../../util/api/RetryFunction.js";
 import { ThrottleFunction } from "../../util/api/ThrottleFunction.js";
-import { PromptMapping } from "../../prompt/PromptMapping.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormat } from "../../prompt/PromptFormat.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 export interface HuggingFaceTextGenerationModelSettings extends TextGenerationModelSettings {
     model: string;
     baseUrl?: string;
@@ -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 }),
  * });
  *
@@ -58,11 +57,8 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
     extractText(response: HuggingFaceTextGenerationResponse): string;
     generateDeltaStreamResponse: undefined;
     extractTextDelta: undefined;
-    mapPrompt<INPUT_PROMPT>(promptMapping: PromptMapping<INPUT_PROMPT, string>): PromptMappingTextGenerationModel<INPUT_PROMPT, string, HuggingFaceTextGenerationResponse, undefined, HuggingFaceTextGenerationModelSettings, this>;
+    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

@@ -3,7 +3,7 @@ import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { callWithRetryAndThrottle } from "../../util/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../util/api/postToApi.js";
 import { failedHuggingFaceCallResponseHandler } from "./HuggingFaceError.js";
-import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingTextGenerationModel.js";
+import { PromptFormatTextGenerationModel } from "../../prompt/PromptFormatTextGenerationModel.js";
 /**
  * Create a text generation model that calls a Hugging Face Inference API Text Generation Task.
  *
@@ -13,7 +13,7 @@ import { PromptMappingTextGenerationModel } from "../../prompt/PromptMappingText
  * 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,
@@ -97,25 +102,15 @@ export class HuggingFaceTextGenerationModel extends AbstractModel {
     extractText(response) {
         return response[0].generated_text;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel({
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel({
             model: this,
-            promptMapping,
+            promptFormat,
         });
     }
     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

@@ -9,7 +9,7 @@ const zod_1 = __importDefault(require("zod"));
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const AsyncQueue_js_1 = require("../../model-function/generate-text/AsyncQueue.cjs");
 const parseEventSourceReadableStream_js_1 = require("../../model-function/generate-text/parseEventSourceReadableStream.cjs");
-const PromptMappingTextGenerationModel_js_1 = require("../../prompt/PromptMappingTextGenerationModel.cjs");
+const PromptFormatTextGenerationModel_js_1 = require("../../prompt/PromptFormatTextGenerationModel.cjs");
 const callWithRetryAndThrottle_js_1 = require("../../util/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../util/api/postToApi.cjs");
 const LlamaCppError_js_1 = require("./LlamaCppError.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,
@@ -76,24 +83,17 @@ class LlamaCppTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     extractTextDelta(fullDelta) {
         return fullDelta.delta;
     }
-    mapPrompt(promptMapping) {
-        return new PromptMappingTextGenerationModel_js_1.PromptMappingTextGenerationModel({
-            model: this.withStopTokens(promptMapping.stopTokens),
-            promptMapping,
+    withPromptFormat(promptFormat) {
+        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
+            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({