modelfusion 0.74.0 → 0.75.0

package/model-provider/openai/TikTokenTokenizer.js

@@ -22,14 +22,14 @@ export class TikTokenTokenizer {
     /**
      * Get a TikToken tokenizer for a specific model or encoding.
      */
-    constructor(options) {
+    constructor(settings) {
         Object.defineProperty(this, "tiktoken", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        this.tiktoken = new Tiktoken(getTiktokenBPE(options.model));
+        this.tiktoken = new Tiktoken(getTiktokenBPE(settings.model));
     }
     async tokenize(text) {
         return this.tiktoken.encode(text);

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

@@ -6,7 +6,8 @@ const AsyncQueue_js_1 = require("../../../util/AsyncQueue.cjs");
 const parseEventSourceStream_js_1 = require("../../../util/streaming/parseEventSourceStream.cjs");
 const parseJSON_js_1 = require("../../../core/schema/parseJSON.cjs");
 const ZodSchema_js_1 = require("../../../core/schema/ZodSchema.cjs");
-const chatResponseStreamEventSchema = new ZodSchema_js_1.ZodSchema(zod_1.z.object({
+const chatCompletionChunkSchema = zod_1.z.object({
+    object: zod_1.z.literal("chat.completion.chunk"),
     id: zod_1.z.string(),
     choices: zod_1.z.array(zod_1.z.object({
         delta: zod_1.z.object({
@@ -44,8 +45,15 @@ const chatResponseStreamEventSchema = new ZodSchema_js_1.ZodSchema(zod_1.z.objec
     created: zod_1.z.number(),
     model: zod_1.z.string(),
     system_fingerprint: zod_1.z.string().optional(),
-    object: zod_1.z.literal("chat.completion.chunk"),
-}));
+});
+const chatResponseStreamEventSchema = new ZodSchema_js_1.ZodSchema(zod_1.z.union([
+    chatCompletionChunkSchema,
+    zod_1.z.object({
+        object: zod_1.z.string().refine((obj) => obj !== "chat.completion.chunk", {
+            message: "Object must not be 'chat.completion.chunk'",
+        }),
+    }),
+]));
 async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaValue) {
     const queue = new AsyncQueue_js_1.AsyncQueue();
     const streamDelta = [];
@@ -68,12 +76,19 @@ async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaValue) {
                         type: "error",
                         error: parseResult.error,
                     });
-                    queue.close();
-                    return;
+                    // Note: the queue is not closed on purpose. Some providers might add additional
+                    // chunks that are not parsable, and ModelFusion should be resilient to that.
+                    continue;
                 }
                 const eventData = parseResult.data;
-                for (let i = 0; i < eventData.choices.length; i++) {
-                    const eventChoice = eventData.choices[i];
+                // ignore objects that are not "chat.completion.chunk" events.
+                // Such additional objects are e.g. sent by Azure OpenAI.
+                if (eventData.object !== "chat.completion.chunk") {
+                    continue;
+                }
+                const completionChunk = eventData;
+                for (let i = 0; i < completionChunk.choices.length; i++) {
+                    const eventChoice = completionChunk.choices[i];
                     const delta = eventChoice.delta;
                     if (streamDelta[i] == null) {
                         streamDelta[i] = {

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

@@ -3,7 +3,8 @@ import { AsyncQueue } from "../../../util/AsyncQueue.js";
 import { parseEventSourceStream } from "../../../util/streaming/parseEventSourceStream.js";
 import { safeParseJSON } from "../../../core/schema/parseJSON.js";
 import { ZodSchema } from "../../../core/schema/ZodSchema.js";
-const chatResponseStreamEventSchema = new ZodSchema(z.object({
+const chatCompletionChunkSchema = z.object({
+    object: z.literal("chat.completion.chunk"),
     id: z.string(),
     choices: z.array(z.object({
         delta: z.object({
@@ -41,8 +42,15 @@ const chatResponseStreamEventSchema = new ZodSchema(z.object({
     created: z.number(),
     model: z.string(),
     system_fingerprint: z.string().optional(),
-    object: z.literal("chat.completion.chunk"),
-}));
+});
+const chatResponseStreamEventSchema = new ZodSchema(z.union([
+    chatCompletionChunkSchema,
+    z.object({
+        object: z.string().refine((obj) => obj !== "chat.completion.chunk", {
+            message: "Object must not be 'chat.completion.chunk'",
+        }),
+    }),
+]));
 export async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaValue) {
     const queue = new AsyncQueue();
     const streamDelta = [];
@@ -65,12 +73,19 @@ export async function createOpenAIChatDeltaIterableQueue(stream, extractDeltaVal
                         type: "error",
                         error: parseResult.error,
                     });
-                    queue.close();
-                    return;
+                    // Note: the queue is not closed on purpose. Some providers might add additional
+                    // chunks that are not parsable, and ModelFusion should be resilient to that.
+                    continue;
                 }
                 const eventData = parseResult.data;
-                for (let i = 0; i < eventData.choices.length; i++) {
-                    const eventChoice = eventData.choices[i];
+                // ignore objects that are not "chat.completion.chunk" events.
+                // Such additional objects are e.g. sent by Azure OpenAI.
+                if (eventData.object !== "chat.completion.chunk") {
+                    continue;
+                }
+                const completionChunk = eventData;
+                for (let i = 0; i < completionChunk.choices.length; i++) {
+                    const eventChoice = completionChunk.choices[i];
                     const delta = eventChoice.delta;
                     if (streamDelta[i] == null) {
                         streamDelta[i] = {

package/model-provider/openai/index.cjs

@@ -26,13 +26,14 @@ var __importStar = (this && this.__importStar) || function (mod) {
     return result;
 };
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.OpenAIChatPromptFormat = exports.OpenAIError = void 0;
+exports.OpenAIChatPromptFormat = exports.openai = exports.OpenAIError = void 0;
 __exportStar(require("./AzureOpenAIApiConfiguration.cjs"), exports);
 __exportStar(require("./OpenAIApiConfiguration.cjs"), exports);
 __exportStar(require("./OpenAICompletionModel.cjs"), exports);
 __exportStar(require("./OpenAICostCalculator.cjs"), exports);
 var OpenAIError_js_1 = require("./OpenAIError.cjs");
 Object.defineProperty(exports, "OpenAIError", { enumerable: true, get: function () { return OpenAIError_js_1.OpenAIError; } });
+exports.openai = __importStar(require("./OpenAIFacade.cjs"));
 __exportStar(require("./OpenAIImageGenerationModel.cjs"), exports);
 __exportStar(require("./OpenAISpeechModel.cjs"), exports);
 __exportStar(require("./OpenAITextEmbeddingModel.cjs"), exports);

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

@@ -3,6 +3,7 @@ export * from "./OpenAIApiConfiguration.js";
 export * from "./OpenAICompletionModel.js";
 export * from "./OpenAICostCalculator.js";
 export { OpenAIError, OpenAIErrorData } from "./OpenAIError.js";
+export * as openai from "./OpenAIFacade.js";
 export * from "./OpenAIImageGenerationModel.js";
 export * from "./OpenAISpeechModel.js";
 export * from "./OpenAITextEmbeddingModel.js";

package/model-provider/openai/index.js

@@ -3,6 +3,7 @@ export * from "./OpenAIApiConfiguration.js";
 export * from "./OpenAICompletionModel.js";
 export * from "./OpenAICostCalculator.js";
 export { OpenAIError } from "./OpenAIError.js";
+export * as openai from "./OpenAIFacade.js";
 export * from "./OpenAIImageGenerationModel.js";
 export * from "./OpenAISpeechModel.js";
 export * from "./OpenAITextEmbeddingModel.js";

package/model-provider/stability/StabilityFacade.cjs

@@ -0,0 +1,32 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.ImageGenerator = void 0;
+const StabilityImageGenerationModel_js_1 = require("./StabilityImageGenerationModel.cjs");
+/**
+ * Create an image generation model that calls the Stability AI image generation API.
+ *
+ * @see https://api.stability.ai/docs#tag/v1generation/operation/textToImage
+ *
+ * @example
+ * const image = await generateImage(
+ *   stability.ImageGenerator({
+ *     model: "stable-diffusion-512-v2-1",
+ *     cfgScale: 7,
+ *     clipGuidancePreset: "FAST_BLUE",
+ *     height: 512,
+ *     width: 512,
+ *     samples: 1,
+ *     steps: 30,
+ *   })
+ *   [
+ *     { text: "the wicked witch of the west" },
+ *     { text: "style of early 19th century painting", weight: 0.5 },
+ *   ]
+ * );
+ *
+ * @returns A new instance of {@link StabilityImageGenerationModel}.
+ */
+function ImageGenerator(settings) {
+    return new StabilityImageGenerationModel_js_1.StabilityImageGenerationModel(settings);
+}
+exports.ImageGenerator = ImageGenerator;

package/model-provider/stability/StabilityFacade.d.ts

@@ -0,0 +1,26 @@
+import { StabilityImageGenerationModel, StabilityImageGenerationSettings } from "./StabilityImageGenerationModel.js";
+/**
+ * Create an image generation model that calls the Stability AI image generation API.
+ *
+ * @see https://api.stability.ai/docs#tag/v1generation/operation/textToImage
+ *
+ * @example
+ * const image = await generateImage(
+ *   stability.ImageGenerator({
+ *     model: "stable-diffusion-512-v2-1",
+ *     cfgScale: 7,
+ *     clipGuidancePreset: "FAST_BLUE",
+ *     height: 512,
+ *     width: 512,
+ *     samples: 1,
+ *     steps: 30,
+ *   })
+ *   [
+ *     { text: "the wicked witch of the west" },
+ *     { text: "style of early 19th century painting", weight: 0.5 },
+ *   ]
+ * );
+ *
+ * @returns A new instance of {@link StabilityImageGenerationModel}.
+ */
+export declare function ImageGenerator(settings: StabilityImageGenerationSettings): StabilityImageGenerationModel;

package/model-provider/stability/StabilityFacade.js

@@ -0,0 +1,28 @@
+import { StabilityImageGenerationModel, } from "./StabilityImageGenerationModel.js";
+/**
+ * Create an image generation model that calls the Stability AI image generation API.
+ *
+ * @see https://api.stability.ai/docs#tag/v1generation/operation/textToImage
+ *
+ * @example
+ * const image = await generateImage(
+ *   stability.ImageGenerator({
+ *     model: "stable-diffusion-512-v2-1",
+ *     cfgScale: 7,
+ *     clipGuidancePreset: "FAST_BLUE",
+ *     height: 512,
+ *     width: 512,
+ *     samples: 1,
+ *     steps: 30,
+ *   })
+ *   [
+ *     { text: "the wicked witch of the west" },
+ *     { text: "style of early 19th century painting", weight: 0.5 },
+ *   ]
+ * );
+ *
+ * @returns A new instance of {@link StabilityImageGenerationModel}.
+ */
+export function ImageGenerator(settings) {
+    return new StabilityImageGenerationModel(settings);
+}

package/model-provider/stability/index.cjs

@@ -10,13 +10,26 @@ var __createBinding = (this && this.__createBinding) || (Object.create ? (functi
     if (k2 === undefined) k2 = k;
     o[k2] = m[k];
 }));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
 var __exportStar = (this && this.__exportStar) || function(m, exports) {
     for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
 };
+var __importStar = (this && this.__importStar) || function (mod) {
+    if (mod && mod.__esModule) return mod;
+    var result = {};
+    if (mod != null) for (var k in mod) if (k !== "default" && Object.prototype.hasOwnProperty.call(mod, k)) __createBinding(result, mod, k);
+    __setModuleDefault(result, mod);
+    return result;
+};
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.StabilityError = void 0;
+exports.stability = exports.StabilityError = void 0;
 __exportStar(require("./StabilityApiConfiguration.cjs"), exports);
 var StabilityError_js_1 = require("./StabilityError.cjs");
 Object.defineProperty(exports, "StabilityError", { enumerable: true, get: function () { return StabilityError_js_1.StabilityError; } });
+exports.stability = __importStar(require("./StabilityFacade.cjs"));
 __exportStar(require("./StabilityImageGenerationModel.cjs"), exports);
 __exportStar(require("./StabilityImageGenerationPrompt.cjs"), exports);

package/model-provider/stability/index.d.ts

@@ -1,4 +1,5 @@
 export * from "./StabilityApiConfiguration.js";
 export { StabilityError, StabilityErrorData } from "./StabilityError.js";
+export * as stability from "./StabilityFacade.js";
 export * from "./StabilityImageGenerationModel.js";
 export * from "./StabilityImageGenerationPrompt.js";

package/model-provider/stability/index.js

@@ -1,4 +1,5 @@
 export * from "./StabilityApiConfiguration.js";
 export { StabilityError } from "./StabilityError.js";
+export * as stability from "./StabilityFacade.js";
 export * from "./StabilityImageGenerationModel.js";
 export * from "./StabilityImageGenerationPrompt.js";

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "modelfusion",
   "description": "The TypeScript library for building multi-modal AI applications.",
-  "version": "0.74.0",
+  "version": "0.75.0",
   "author": "Lars Grammel",
   "license": "MIT",
   "keywords": [

package/model-provider/huggingface/HuggingFaceImageDescriptionModel.cjs

@@ -1,94 +0,0 @@
-"use strict";
-Object.defineProperty(exports, "__esModule", { value: true });
-exports.HuggingFaceImageDescriptionModel = void 0;
-const zod_1 = require("zod");
-const callWithRetryAndThrottle_js_1 = require("../../core/api/callWithRetryAndThrottle.cjs");
-const postToApi_js_1 = require("../../core/api/postToApi.cjs");
-const AbstractModel_js_1 = require("../../model-function/AbstractModel.cjs");
-const PromptFormatTextGenerationModel_js_1 = require("../../model-function/generate-text/PromptFormatTextGenerationModel.cjs");
-const HuggingFaceApiConfiguration_js_1 = require("./HuggingFaceApiConfiguration.cjs");
-const HuggingFaceError_js_1 = require("./HuggingFaceError.cjs");
-/**
- * Create an image to text model that calls a Hugging Face Image-to-Text Inference API.
- *
- * @see https://huggingface.co/tasks/image-to-text
- */
-class HuggingFaceImageDescriptionModel extends AbstractModel_js_1.AbstractModel {
-    constructor(settings) {
-        super({ settings });
-        Object.defineProperty(this, "provider", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: "huggingface"
-        });
-        Object.defineProperty(this, "contextWindowSize", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-        Object.defineProperty(this, "tokenizer", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-        Object.defineProperty(this, "countPromptTokens", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-    }
-    get modelName() {
-        return this.settings.model;
-    }
-    async callAPI(data, options) {
-        return (0, callWithRetryAndThrottle_js_1.callWithRetryAndThrottle)({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
-            call: async () => callHuggingFaceImageDescriptionAPI({
-                ...this.settings,
-                abortSignal: options?.run?.abortSignal,
-                data,
-            }),
-        });
-    }
-    get settingsForEvent() {
-        return {};
-    }
-    async doGenerateText(data, options) {
-        const response = await this.callAPI(data, options);
-        return {
-            response,
-            text: response[0].generated_text,
-        };
-    }
-    withPromptFormat(promptFormat) {
-        return new PromptFormatTextGenerationModel_js_1.PromptFormatTextGenerationModel({
-            model: this,
-            promptFormat,
-        });
-    }
-    withSettings(additionalSettings) {
-        return new HuggingFaceImageDescriptionModel(Object.assign({}, this.settings, additionalSettings));
-    }
-}
-exports.HuggingFaceImageDescriptionModel = HuggingFaceImageDescriptionModel;
-const huggingFaceImageDescriptionResponseSchema = zod_1.z.array(zod_1.z.object({
-    generated_text: zod_1.z.string(),
-}));
-async function callHuggingFaceImageDescriptionAPI({ api = new HuggingFaceApiConfiguration_js_1.HuggingFaceApiConfiguration(), abortSignal, model, data, }) {
-    return (0, postToApi_js_1.postToApi)({
-        url: api.assembleUrl(`/${model}`),
-        headers: api.headers,
-        body: {
-            content: data,
-            values: {},
-        },
-        failedResponseHandler: HuggingFaceError_js_1.failedHuggingFaceCallResponseHandler,
-        successfulResponseHandler: (0, postToApi_js_1.createJsonResponseHandler)(huggingFaceImageDescriptionResponseSchema),
-        abortSignal,
-    });
-}

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

@@ -1,44 +0,0 @@
-/// <reference types="node" />
-import { z } from "zod";
-import { FunctionOptions } from "../../core/FunctionOptions.js";
-import { ApiConfiguration } from "../../core/api/ApiConfiguration.js";
-import { AbstractModel } from "../../model-function/AbstractModel.js";
-import { PromptFormatTextGenerationModel } from "../../model-function/generate-text/PromptFormatTextGenerationModel.js";
-import { TextGenerationModel, TextGenerationModelSettings } from "../../model-function/generate-text/TextGenerationModel.js";
-import { TextGenerationPromptFormat } from "../../model-function/generate-text/TextGenerationPromptFormat.js";
-export interface HuggingFaceImageDescriptionModelSettings extends TextGenerationModelSettings {
-    api?: ApiConfiguration;
-    model: string;
-}
-/**
- * Create an image to text model that calls a Hugging Face Image-to-Text Inference API.
- *
- * @see https://huggingface.co/tasks/image-to-text
- */
-export declare class HuggingFaceImageDescriptionModel extends AbstractModel<HuggingFaceImageDescriptionModelSettings> implements TextGenerationModel<Buffer, HuggingFaceImageDescriptionModelSettings> {
-    constructor(settings: HuggingFaceImageDescriptionModelSettings);
-    readonly provider = "huggingface";
-    get modelName(): string;
-    callAPI(data: Buffer, options?: FunctionOptions): Promise<HuggingFaceImageDescriptionResponse>;
-    get settingsForEvent(): Partial<HuggingFaceImageDescriptionModelSettings>;
-    readonly contextWindowSize: undefined;
-    readonly tokenizer: undefined;
-    readonly countPromptTokens: undefined;
-    doGenerateText(data: Buffer, options?: FunctionOptions): Promise<{
-        response: {
-            generated_text: string;
-        }[];
-        text: string;
-    }>;
-    withPromptFormat<INPUT_PROMPT>(promptFormat: TextGenerationPromptFormat<INPUT_PROMPT, Buffer>): PromptFormatTextGenerationModel<INPUT_PROMPT, Buffer, HuggingFaceImageDescriptionModelSettings, this>;
-    withSettings(additionalSettings: Partial<HuggingFaceImageDescriptionModelSettings>): this;
-}
-declare const huggingFaceImageDescriptionResponseSchema: z.ZodArray<z.ZodObject<{
-    generated_text: z.ZodString;
-}, "strip", z.ZodTypeAny, {
-    generated_text: string;
-}, {
-    generated_text: string;
-}>, "many">;
-export type HuggingFaceImageDescriptionResponse = z.infer<typeof huggingFaceImageDescriptionResponseSchema>;
-export {};

package/model-provider/huggingface/HuggingFaceImageDescriptionModel.js

@@ -1,90 +0,0 @@
-import { z } from "zod";
-import { callWithRetryAndThrottle } from "../../core/api/callWithRetryAndThrottle.js";
-import { createJsonResponseHandler, postToApi, } from "../../core/api/postToApi.js";
-import { AbstractModel } from "../../model-function/AbstractModel.js";
-import { PromptFormatTextGenerationModel } from "../../model-function/generate-text/PromptFormatTextGenerationModel.js";
-import { HuggingFaceApiConfiguration } from "./HuggingFaceApiConfiguration.js";
-import { failedHuggingFaceCallResponseHandler } from "./HuggingFaceError.js";
-/**
- * Create an image to text model that calls a Hugging Face Image-to-Text Inference API.
- *
- * @see https://huggingface.co/tasks/image-to-text
- */
-export class HuggingFaceImageDescriptionModel extends AbstractModel {
-    constructor(settings) {
-        super({ settings });
-        Object.defineProperty(this, "provider", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: "huggingface"
-        });
-        Object.defineProperty(this, "contextWindowSize", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-        Object.defineProperty(this, "tokenizer", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-        Object.defineProperty(this, "countPromptTokens", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: undefined
-        });
-    }
-    get modelName() {
-        return this.settings.model;
-    }
-    async callAPI(data, options) {
-        return callWithRetryAndThrottle({
-            retry: this.settings.api?.retry,
-            throttle: this.settings.api?.throttle,
-            call: async () => callHuggingFaceImageDescriptionAPI({
-                ...this.settings,
-                abortSignal: options?.run?.abortSignal,
-                data,
-            }),
-        });
-    }
-    get settingsForEvent() {
-        return {};
-    }
-    async doGenerateText(data, options) {
-        const response = await this.callAPI(data, options);
-        return {
-            response,
-            text: response[0].generated_text,
-        };
-    }
-    withPromptFormat(promptFormat) {
-        return new PromptFormatTextGenerationModel({
-            model: this,
-            promptFormat,
-        });
-    }
-    withSettings(additionalSettings) {
-        return new HuggingFaceImageDescriptionModel(Object.assign({}, this.settings, additionalSettings));
-    }
-}
-const huggingFaceImageDescriptionResponseSchema = z.array(z.object({
-    generated_text: z.string(),
-}));
-async function callHuggingFaceImageDescriptionAPI({ api = new HuggingFaceApiConfiguration(), abortSignal, model, data, }) {
-    return postToApi({
-        url: api.assembleUrl(`/${model}`),
-        headers: api.headers,
-        body: {
-            content: data,
-            values: {},
-        },
-        failedResponseHandler: failedHuggingFaceCallResponseHandler,
-        successfulResponseHandler: createJsonResponseHandler(huggingFaceImageDescriptionResponseSchema),
-        abortSignal,
-    });
-}