modelfusion 0.102.0 → 0.103.0

package/CHANGELOG.md

@@ -1,5 +1,25 @@
 # Changelog
 
+## v0.103.0 - 2023-12-23
+
+### Added
+
+- `finishReason` support for `generateText`.
+
+  The finish reason can be `stop` (the model stopped because it generated a stop sequence), `length` (the model stopped because it generated the maximum number of tokens), `content-filter` (the model stopped because the content filter detected a violation), `tool-calls` (the model stopped because it triggered a tool call), `error` (the model stopped because of an error), `other` (the model stopped for another reason), or `unknown` (the model stop reason is not know or the model does not support finish reasons).
+
+  You can extract it from the full response when using `fullResponse: true`:
+
+  ```ts
+  const { text, finishReason } = await generateText(
+    openai
+      .ChatTextGenerator({ model: "gpt-3.5-turbo", maxGenerationTokens: 200 })
+      .withTextPrompt(),
+    "Write a short story about a robot learning to love:",
+    { fullResponse: true }
+  );
+  ```
+
 ## v0.102.0 - 2023-12-22
 
 ### Added

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

@@ -19,7 +19,7 @@ export declare class PromptTemplateTextGenerationModel<PROMPT, MODEL_PROMPT, SET
     get countPromptTokens(): MODEL["countPromptTokens"] extends undefined ? undefined : (prompt: PROMPT) => PromiseLike<number>;
     doGenerateTexts(prompt: PROMPT, options?: FunctionOptions): PromiseLike<{
         response: unknown;
-        texts: string[];
+        textGenerationResults: import("./TextGenerationResult.js").TextGenerationResult[];
         usage?: {
             promptTokens: number;
             completionTokens: number;

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

@@ -1,2 +1,9 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.textGenerationModelProperties = void 0;
+exports.textGenerationModelProperties = [
+    "maxGenerationTokens",
+    "stopSequences",
+    "numberOfGenerations",
+    "trimWhitespace",
+];

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

@@ -3,6 +3,8 @@ import { Delta } from "../Delta.js";
 import { Model, ModelSettings } from "../Model.js";
 import { BasicTokenizer, FullTokenizer } from "../tokenize-text/Tokenizer.js";
 import { TextGenerationPromptTemplate } from "./TextGenerationPromptTemplate.js";
+import { TextGenerationResult } from "./TextGenerationResult.js";
+export declare const textGenerationModelProperties: readonly ["maxGenerationTokens", "stopSequences", "numberOfGenerations", "trimWhitespace"];
 export interface TextGenerationModelSettings extends ModelSettings {
     /**
      * Specifies the maximum number of tokens (words, punctuation, parts of words) that the model can generate in a single response.
@@ -62,7 +64,7 @@ export interface TextGenerationModel<PROMPT, SETTINGS extends TextGenerationMode
     readonly countPromptTokens: ((prompt: PROMPT) => PromiseLike<number>) | undefined;
     doGenerateTexts(prompt: PROMPT, options?: FunctionOptions): PromiseLike<{
         response: unknown;
-        texts: string[];
+        textGenerationResults: TextGenerationResult[];
         usage?: {
             promptTokens: number;
             completionTokens: number;

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

@@ -1 +1,6 @@
-export {};
+export const textGenerationModelProperties = [
+    "maxGenerationTokens",
+    "stopSequences",
+    "numberOfGenerations",
+    "trimWhitespace",
+];

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

@@ -0,0 +1,2 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });

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

@@ -0,0 +1,11 @@
+export type TextGenerationResult = {
+    /**
+     * The generated text.
+     */
+    text: string;
+    /**
+     * The reason why the generation stopped.
+     */
+    finishReason: TextGenerationFinishReason;
+};
+export type TextGenerationFinishReason = "stop" | "length" | "content-filter" | "tool-calls" | "error" | "other" | "unknown";

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

@@ -0,0 +1 @@
+export {};

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

@@ -11,25 +11,30 @@ async function generateText(model, prompt, options) {
         generateResponse: async (options) => {
             const result = await model.doGenerateTexts(prompt, options);
             const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
-            const texts = shouldTrimWhitespace
-                ? result.texts.map((text) => text.trim())
-                : result.texts;
+            const textGenerationResults = shouldTrimWhitespace
+                ? result.textGenerationResults.map((textGeneration) => ({
+                    text: textGeneration.text.trim(),
+                    finishReason: textGeneration.finishReason,
+                }))
+                : result.textGenerationResults;
             return {
                 response: result.response,
-                extractedValue: texts,
+                extractedValue: textGenerationResults,
                 usage: result.usage,
             };
         },
     });
-    const texts = fullResponse.value;
-    const text = texts[0];
+    const textGenerationResults = fullResponse.value;
+    const firstResult = textGenerationResults[0];
     return options?.fullResponse
         ? {
-            text,
-            texts,
+            text: firstResult.text,
+            finishReason: firstResult.finishReason,
+            texts: textGenerationResults.map((textGeneration) => textGeneration.text),
+            textGenerationResults,
             response: fullResponse.response,
             metadata: fullResponse.metadata,
         }
-        : text;
+        : firstResult.text;
 }
 exports.generateText = generateText;

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

@@ -1,6 +1,7 @@
 import { FunctionOptions } from "../../core/FunctionOptions.js";
 import { ModelCallMetadata } from "../ModelCallMetadata.js";
 import { TextGenerationModel, TextGenerationModelSettings } from "./TextGenerationModel.js";
+import { TextGenerationFinishReason, TextGenerationResult } from "./TextGenerationResult.js";
 /**
  * Generate text for a prompt and return it as a string.
  *
@@ -29,7 +30,9 @@ export declare function generateText<PROMPT>(model: TextGenerationModel<PROMPT,
     fullResponse: true;
 }): Promise<{
     text: string;
+    finishReason: TextGenerationFinishReason;
     texts: string[];
+    textGenerationResults: TextGenerationResult[];
     response: unknown;
     metadata: ModelCallMetadata;
 }>;

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

@@ -8,24 +8,29 @@ export async function generateText(model, prompt, options) {
         generateResponse: async (options) => {
             const result = await model.doGenerateTexts(prompt, options);
             const shouldTrimWhitespace = model.settings.trimWhitespace ?? true;
-            const texts = shouldTrimWhitespace
-                ? result.texts.map((text) => text.trim())
-                : result.texts;
+            const textGenerationResults = shouldTrimWhitespace
+                ? result.textGenerationResults.map((textGeneration) => ({
+                    text: textGeneration.text.trim(),
+                    finishReason: textGeneration.finishReason,
+                }))
+                : result.textGenerationResults;
             return {
                 response: result.response,
-                extractedValue: texts,
+                extractedValue: textGenerationResults,
                 usage: result.usage,
             };
         },
     });
-    const texts = fullResponse.value;
-    const text = texts[0];
+    const textGenerationResults = fullResponse.value;
+    const firstResult = textGenerationResults[0];
     return options?.fullResponse
         ? {
-            text,
-            texts,
+            text: firstResult.text,
+            finishReason: firstResult.finishReason,
+            texts: textGenerationResults.map((textGeneration) => textGeneration.text),
+            textGenerationResults,
             response: fullResponse.response,
             metadata: fullResponse.metadata,
         }
-        : text;
+        : firstResult.text;
 }

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

@@ -19,6 +19,7 @@ __exportStar(require("./PromptTemplateTextStreamingModel.cjs"), exports);
 __exportStar(require("./TextGenerationEvent.cjs"), exports);
 __exportStar(require("./TextGenerationModel.cjs"), exports);
 __exportStar(require("./TextGenerationPromptTemplate.cjs"), exports);
+__exportStar(require("./TextGenerationResult.cjs"), exports);
 __exportStar(require("./generateText.cjs"), exports);
 __exportStar(require("./prompt-template/index.cjs"), exports);
 __exportStar(require("./streamText.cjs"), exports);

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

@@ -3,6 +3,7 @@ export * from "./PromptTemplateTextStreamingModel.js";
 export * from "./TextGenerationEvent.js";
 export * from "./TextGenerationModel.js";
 export * from "./TextGenerationPromptTemplate.js";
+export * from "./TextGenerationResult.js";
 export * from "./generateText.js";
 export * from "./prompt-template/index.js";
 export * from "./streamText.js";

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

@@ -3,6 +3,7 @@ export * from "./PromptTemplateTextStreamingModel.js";
 export * from "./TextGenerationEvent.js";
 export * from "./TextGenerationModel.js";
 export * from "./TextGenerationPromptTemplate.js";
+export * from "./TextGenerationResult.js";
 export * from "./generateText.js";
 export * from "./prompt-template/index.js";
 export * from "./streamText.js";

package/model-provider/anthropic/AnthropicTextGenerationModel.cjs

@@ -4,12 +4,13 @@ exports.AnthropicTextGenerationResponseFormat = exports.AnthropicTextGenerationM
 const zod_1 = require("zod");
 const callWithRetryAndThrottle_js_1 = require("../../core/api/callWithRetryAndThrottle.cjs");
 const postToApi_js_1 = require("../../core/api/postToApi.cjs");
-const parseEventSourceStream_js_1 = require("../../util/streaming/parseEventSourceStream.cjs");
+const ZodSchema_js_1 = require("../../core/schema/ZodSchema.cjs");
+const parseJSON_js_1 = require("../../core/schema/parseJSON.cjs");
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const PromptTemplateTextStreamingModel_js_1 = require("../../model-function/generate-text/PromptTemplateTextStreamingModel.cjs");
+const TextGenerationModel_js_1 = require("../../model-function/generate-text/TextGenerationModel.cjs");
 const AsyncQueue_js_1 = require("../../util/AsyncQueue.cjs");
-const ZodSchema_js_1 = require("../../core/schema/ZodSchema.cjs");
-const parseJSON_js_1 = require("../../core/schema/parseJSON.cjs");
+const parseEventSourceStream_js_1 = require("../../util/streaming/parseEventSourceStream.cjs");
 const AnthropicApiConfiguration_js_1 = require("./AnthropicApiConfiguration.cjs");
 const AnthropicError_js_1 = require("./AnthropicError.cjs");
 const AnthropicPromptTemplate_js_1 = require("./AnthropicPromptTemplate.cjs");
@@ -74,8 +75,8 @@ class AnthropicTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         const abortSignal = options.run?.abortSignal;
         const userId = this.settings.userId;
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
+            retry: api.retry,
+            throttle: api.throttle,
             call: async () => {
                 return (0, postToApi_js_1.postJsonToApi)({
                     url: api.assembleUrl(`/complete`),
@@ -100,8 +101,7 @@ class AnthropicTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxGenerationTokens",
-            "stopSequences",
+            ...TextGenerationModel_js_1.textGenerationModelProperties,
             "temperature",
             "topK",
             "topP",
@@ -116,9 +116,24 @@ class AnthropicTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         });
         return {
             response,
-            texts: [response.completion],
+            textGenerationResults: [
+                {
+                    text: response.completion,
+                    finishReason: this.translateFinishReason(response.stop_reason),
+                },
+            ],
         };
     }
+    translateFinishReason(finishReason) {
+        switch (finishReason) {
+            case "stop_sequence":
+                return "stop";
+            case "max_tokens":
+                return "length";
+            default:
+                return "unknown";
+        }
+    }
     doStreamText(prompt, options) {
         return this.callAPI(prompt, {
             ...options,

package/model-provider/anthropic/AnthropicTextGenerationModel.d.ts

@@ -7,6 +7,7 @@ import { Delta } from "../../model-function/Delta.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
 import { TextGenerationModelSettings, TextStreamingModel } from "../../model-function/generate-text/TextGenerationModel.js";
 import { TextGenerationPromptTemplate } from "../../model-function/generate-text/TextGenerationPromptTemplate.js";
+import { TextGenerationFinishReason } from "../../model-function/generate-text/TextGenerationResult.js";
 export declare const ANTHROPIC_TEXT_GENERATION_MODELS: {
     "claude-instant-1": {
         contextWindowSize: number;
@@ -55,8 +56,12 @@ export declare class AnthropicTextGenerationModel extends AbstractModel<Anthropi
             completion: string;
             stop_reason: string;
         };
-        texts: string[];
+        textGenerationResults: {
+            text: string;
+            finishReason: TextGenerationFinishReason;
+        }[];
     }>;
+    private translateFinishReason;
     doStreamText(prompt: string, options?: FunctionOptions): Promise<AsyncIterable<Delta<string>>>;
     /**
      * Returns this model with a text prompt template.

package/model-provider/anthropic/AnthropicTextGenerationModel.js

@@ -1,15 +1,16 @@
 import { z } from "zod";
 import { callWithRetryAndThrottle } from "../../core/api/callWithRetryAndThrottle.js";
 import { createJsonResponseHandler, postJsonToApi, } from "../../core/api/postToApi.js";
-import { parseEventSourceStream } from "../../util/streaming/parseEventSourceStream.js";
+import { ZodSchema } from "../../core/schema/ZodSchema.js";
+import { parseJSON } from "../../core/schema/parseJSON.js";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
+import { textGenerationModelProperties, } from "../../model-function/generate-text/TextGenerationModel.js";
 import { AsyncQueue } from "../../util/AsyncQueue.js";
-import { ZodSchema } from "../../core/schema/ZodSchema.js";
-import { parseJSON } from "../../core/schema/parseJSON.js";
+import { parseEventSourceStream } from "../../util/streaming/parseEventSourceStream.js";
 import { AnthropicApiConfiguration } from "./AnthropicApiConfiguration.js";
 import { failedAnthropicCallResponseHandler } from "./AnthropicError.js";
-import { instruction, chat, text } from "./AnthropicPromptTemplate.js";
+import { chat, instruction, text } from "./AnthropicPromptTemplate.js";
 export const ANTHROPIC_TEXT_GENERATION_MODELS = {
     "claude-instant-1": {
         contextWindowSize: 100000,
@@ -71,8 +72,8 @@ export class AnthropicTextGenerationModel extends AbstractModel {
         const abortSignal = options.run?.abortSignal;
         const userId = this.settings.userId;
         return callWithRetryAndThrottle({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
+            retry: api.retry,
+            throttle: api.throttle,
             call: async () => {
                 return postJsonToApi({
                     url: api.assembleUrl(`/complete`),
@@ -97,8 +98,7 @@ export class AnthropicTextGenerationModel extends AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxGenerationTokens",
-            "stopSequences",
+            ...textGenerationModelProperties,
             "temperature",
             "topK",
             "topP",
@@ -113,9 +113,24 @@ export class AnthropicTextGenerationModel extends AbstractModel {
         });
         return {
             response,
-            texts: [response.completion],
+            textGenerationResults: [
+                {
+                    text: response.completion,
+                    finishReason: this.translateFinishReason(response.stop_reason),
+                },
+            ],
         };
     }
+    translateFinishReason(finishReason) {
+        switch (finishReason) {
+            case "stop_sequence":
+                return "stop";
+            case "max_tokens":
+                return "length";
+            default:
+                return "unknown";
+        }
+    }
     doStreamText(prompt, options) {
         return this.callAPI(prompt, {
             ...options,

package/model-provider/cohere/CohereTextGenerationModel.cjs

@@ -7,6 +7,7 @@ const postToApi_js_1 = require("../../core/api/postToApi.cjs");
 const ZodSchema_js_1 = require("../../core/schema/ZodSchema.cjs");
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const PromptTemplateTextStreamingModel_js_1 = require("../../model-function/generate-text/PromptTemplateTextStreamingModel.cjs");
+const TextGenerationModel_js_1 = require("../../model-function/generate-text/TextGenerationModel.cjs");
 const TextPromptTemplate_js_1 = require("../../model-function/generate-text/prompt-template/TextPromptTemplate.cjs");
 const countTokens_js_1 = require("../../model-function/tokenize-text/countTokens.cjs");
 const AsyncQueue_js_1 = require("../../util/AsyncQueue.cjs");
@@ -84,8 +85,8 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         const responseFormat = options.responseFormat;
         const abortSignal = options.run?.abortSignal;
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
+            retry: api.retry,
+            throttle: api.throttle,
             call: async () => {
                 return (0, postToApi_js_1.postJsonToApi)({
                     url: api.assembleUrl(`/generate`),
@@ -116,9 +117,7 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxGenerationTokens",
-            "stopSequences",
-            "numberOfGenerations",
+            ...TextGenerationModel_js_1.textGenerationModelProperties,
             "temperature",
             "k",
             "p",
@@ -138,9 +137,26 @@ class CohereTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         });
         return {
             response,
-            texts: response.generations.map((generation) => generation.text),
+            textGenerationResults: response.generations.map((generation) => ({
+                text: generation.text,
+                finishReason: this.translateFinishReason(generation.finish_reason),
+            })),
         };
     }
+    translateFinishReason(finishReason) {
+        switch (finishReason) {
+            case "COMPLETE":
+                return "stop";
+            case "MAX_TOKENS":
+                return "length";
+            case "ERROR_TOXIC":
+                return "content-filter";
+            case "ERROR":
+                return "error";
+            default:
+                return "unknown";
+        }
+    }
     doStreamText(prompt, options) {
         return this.callAPI(prompt, {
             ...options,

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

@@ -7,6 +7,7 @@ import { Delta } from "../../model-function/Delta.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
 import { TextGenerationModelSettings, TextStreamingModel } from "../../model-function/generate-text/TextGenerationModel.js";
 import { TextGenerationPromptTemplate } from "../../model-function/generate-text/TextGenerationPromptTemplate.js";
+import { TextGenerationFinishReason } from "../../model-function/generate-text/TextGenerationResult.js";
 import { CohereTokenizer } from "./CohereTokenizer.js";
 export declare const COHERE_TEXT_GENERATION_MODELS: {
     command: {
@@ -79,8 +80,12 @@ export declare class CohereTextGenerationModel extends AbstractModel<CohereTextG
                 };
             } | undefined;
         };
-        texts: string[];
+        textGenerationResults: {
+            text: string;
+            finishReason: TextGenerationFinishReason;
+        }[];
     }>;
+    private translateFinishReason;
     doStreamText(prompt: string, options?: FunctionOptions): Promise<AsyncIterable<Delta<string>>>;
     extractTextDelta(fullDelta: CohereTextGenerationDelta): string | undefined;
     /**

package/model-provider/cohere/CohereTextGenerationModel.js

@@ -4,6 +4,7 @@ import { createJsonResponseHandler, postJsonToApi, } from "../../core/api/postTo
 import { ZodSchema } from "../../core/schema/ZodSchema.js";
 import { AbstractModel } from "../../model-function/AbstractModel.js";
 import { PromptTemplateTextStreamingModel } from "../../model-function/generate-text/PromptTemplateTextStreamingModel.js";
+import { textGenerationModelProperties, } from "../../model-function/generate-text/TextGenerationModel.js";
 import { chat, instruction, } from "../../model-function/generate-text/prompt-template/TextPromptTemplate.js";
 import { countTokens } from "../../model-function/tokenize-text/countTokens.js";
 import { AsyncQueue } from "../../util/AsyncQueue.js";
@@ -81,8 +82,8 @@ export class CohereTextGenerationModel extends AbstractModel {
         const responseFormat = options.responseFormat;
         const abortSignal = options.run?.abortSignal;
         return callWithRetryAndThrottle({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
+            retry: api.retry,
+            throttle: api.throttle,
             call: async () => {
                 return postJsonToApi({
                     url: api.assembleUrl(`/generate`),
@@ -113,9 +114,7 @@ export class CohereTextGenerationModel extends AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "maxGenerationTokens",
-            "stopSequences",
-            "numberOfGenerations",
+            ...textGenerationModelProperties,
             "temperature",
             "k",
             "p",
@@ -135,9 +134,26 @@ export class CohereTextGenerationModel extends AbstractModel {
         });
         return {
             response,
-            texts: response.generations.map((generation) => generation.text),
+            textGenerationResults: response.generations.map((generation) => ({
+                text: generation.text,
+                finishReason: this.translateFinishReason(generation.finish_reason),
+            })),
         };
     }
+    translateFinishReason(finishReason) {
+        switch (finishReason) {
+            case "COMPLETE":
+                return "stop";
+            case "MAX_TOKENS":
+                return "length";
+            case "ERROR_TOXIC":
+                return "content-filter";
+            case "ERROR":
+                return "error";
+            default:
+                return "unknown";
+        }
+    }
     doStreamText(prompt, options) {
         return this.callAPI(prompt, {
             ...options,

package/model-provider/elevenlabs/ElevenLabsSpeechModel.cjs

@@ -101,7 +101,7 @@ class ElevenLabsSpeechModel extends AbstractModel_js_1.AbstractModel {
                 // The JS WebSocket API does not support authorization headers, so we send the API key in the BOS message.
                 // See https://stackoverflow.com/questions/4361173/http-headers-in-websockets-client-api
                 xi_api_key: api.apiKey,
-                text: " ",
+                text: " ", // first message
                 voice_settings: toApiVoiceSettings(this.settings.voiceSettings),
                 generation_config: toGenerationConfig(this.settings.generationConfig),
             }));
@@ -125,7 +125,7 @@ class ElevenLabsSpeechModel extends AbstractModel_js_1.AbstractModel {
             // send remaining text:
             if (textBuffer.length > 0) {
                 socket.send(JSON.stringify({
-                    text: `${textBuffer} `,
+                    text: `${textBuffer} `, // append space
                     try_trigger_generation: true,
                 }));
             }

package/model-provider/elevenlabs/ElevenLabsSpeechModel.js

@@ -98,7 +98,7 @@ export class ElevenLabsSpeechModel extends AbstractModel {
                 // The JS WebSocket API does not support authorization headers, so we send the API key in the BOS message.
                 // See https://stackoverflow.com/questions/4361173/http-headers-in-websockets-client-api
                 xi_api_key: api.apiKey,
-                text: " ",
+                text: " ", // first message
                 voice_settings: toApiVoiceSettings(this.settings.voiceSettings),
                 generation_config: toGenerationConfig(this.settings.generationConfig),
             }));
@@ -122,7 +122,7 @@ export class ElevenLabsSpeechModel extends AbstractModel {
             // send remaining text:
             if (textBuffer.length > 0) {
                 socket.send(JSON.stringify({
-                    text: `${textBuffer} `,
+                    text: `${textBuffer} `, // append space
                     try_trigger_generation: true,
                 }));
             }

package/model-provider/huggingface/HuggingFaceTextGenerationModel.cjs

@@ -6,6 +6,7 @@ const callWithRetryAndThrottle_js_1 = require("../../core/api/callWithRetryAndTh
 const postToApi_js_1 = require("../../core/api/postToApi.cjs");
 const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
 const PromptTemplateTextGenerationModel_js_1 = require("../../model-function/generate-text/PromptTemplateTextGenerationModel.cjs");
+const TextGenerationModel_js_1 = require("../../model-function/generate-text/TextGenerationModel.cjs");
 const HuggingFaceApiConfiguration_js_1 = require("./HuggingFaceApiConfiguration.cjs");
 const HuggingFaceError_js_1 = require("./HuggingFaceError.cjs");
 /**
@@ -61,8 +62,8 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         const api = this.settings.api ?? new HuggingFaceApiConfiguration_js_1.HuggingFaceApiConfiguration();
         const abortSignal = options?.run?.abortSignal;
         return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
+            retry: api.retry,
+            throttle: api.throttle,
             call: async () => {
                 return (0, postToApi_js_1.postJsonToApi)({
                     url: api.assembleUrl(`/${this.settings.model}`),
@@ -91,16 +92,13 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
     }
     get settingsForEvent() {
         const eventSettingProperties = [
-            "stopSequences",
-            "maxGenerationTokens",
-            "numberOfGenerations",
+            ...TextGenerationModel_js_1.textGenerationModelProperties,
             "topK",
             "topP",
             "temperature",
             "repetitionPenalty",
             "maxTime",
             "doSample",
-            "options",
         ];
         return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
     }
@@ -108,12 +106,15 @@ class HuggingFaceTextGenerationModel extends AbstractModel_js_1.AbstractModel {
         const response = await this.callAPI(prompt, options);
         return {
             response,
-            texts: response.map((response) => response.generated_text),
+            textGenerationResults: response.map((response) => ({
+                text: response.generated_text,
+                finishReason: "unknown",
+            })),
         };
     }
     withPromptTemplate(promptTemplate) {
         return new PromptTemplateTextGenerationModel_js_1.PromptTemplateTextGenerationModel({
-            model: this,
+            model: this, // stop tokens are not supported by this model
             promptTemplate,
         });
     }

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

@@ -14,10 +14,6 @@ export interface HuggingFaceTextGenerationModelSettings extends TextGenerationMo
     repetitionPenalty?: number;
     maxTime?: number;
     doSample?: boolean;
-    options?: {
-        useCache?: boolean;
-        waitForModel?: boolean;
-    };
 }
 /**
  * Create a text generation model that calls a Hugging Face Inference API Text Generation Task.
@@ -50,7 +46,10 @@ export declare class HuggingFaceTextGenerationModel extends AbstractModel<Huggin
         response: {
             generated_text: string;
         }[];
-        texts: string[];
+        textGenerationResults: {
+            text: string;
+            finishReason: "unknown";
+        }[];
     }>;
     withPromptTemplate<INPUT_PROMPT>(promptTemplate: TextGenerationPromptTemplate<INPUT_PROMPT, string>): PromptTemplateTextGenerationModel<INPUT_PROMPT, string, HuggingFaceTextGenerationModelSettings, this>;
     withSettings(additionalSettings: Partial<HuggingFaceTextGenerationModelSettings>): this;