@huggingface/inference 1.7.1 → 1.8.0

package/README.md

@@ -4,8 +4,6 @@ A Typescript powered wrapper for the Hugging Face Inference API. Learn more abou
 
 Check out the [full documentation](https://huggingface.co/docs/huggingface.js/inference/README) or try out a live [interactive notebook](https://observablehq.com/@huggingface/hello-huggingface-js-inference).
 
-
-
 ## Install
 
 ```console
@@ -18,7 +16,7 @@ pnpm add @huggingface/inference
 
 ## Usage
 
-❗**Important note:** Using an API key is optional to get started, however you will be rate limited eventually. Join [Hugging Face](https://huggingface.co/join) and then visit [access tokens](https://huggingface.co/settings/tokens) to generate your API key for **free**. 
+❗**Important note:** Using an API key is optional to get started, however you will be rate limited eventually. Join [Hugging Face](https://huggingface.co/join) and then visit [access tokens](https://huggingface.co/settings/tokens) to generate your API key for **free**.
 
 Your API key should be kept private. If you need to protect it in front-end applications, we suggest setting up a proxy server that stores the API key.
 
@@ -154,8 +152,15 @@ await hf.imageSegmentation({
 
 await hf.textToImage({
   inputs: 'award winning high resolution photo of a giant tortoise/((ladybird)) hybrid, [trending on artstation]',
-  negative_prompt: 'blurry',
   model: 'stabilityai/stable-diffusion-2',
+  parameters: {
+    negative_prompt: 'blurry',
+  }
+})
+
+await hf.imageToText({
+  data: readFileSync('test/cats.png'),
+  model: 'nlpconnect/vit-gpt2-image-captioning'
 })
 ```
 
@@ -188,6 +193,7 @@ await hf.textToImage({
 - [x] Object detection
 - [x] Image segmentation
 - [x] Text to image
+- [x] Image to text
 
 ## Running tests
 

package/dist/index.d.ts

@@ -273,12 +273,12 @@ interface TextGenerationStreamReturn {
      * Complete generated text
      * Only available when the generation is finished
      */
-    generated_text?: string;
+    generated_text: string | null;
     /**
      * Generation details
      * Only available when the generation is finished
      */
-    details?: TextGenerationStreamDetails;
+    details: TextGenerationStreamDetails | null;
 }
 type TokenClassificationArgs = Args & {
     /**
@@ -526,12 +526,42 @@ type TextToImageArgs = Args & {
      * The text to generate an image from
      */
     inputs: string;
+    parameters?: {
+        /**
+         * An optional negative prompt for the image generation
+         */
+        negative_prompt?: string;
+        /**
+         * The height in pixels of the generated image
+         */
+        height?: number;
+        /**
+         * The width in pixels of the generated image
+         */
+        width?: number;
+        /**
+         * The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference.
+         */
+        num_inference_steps?: number;
+        /**
+         * Guidance scale: Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality.
+         */
+        guidance_scale?: number;
+    };
+};
+type TextToImageReturn = Blob;
+type ImageToTextArgs = Args & {
     /**
-     * An optional negative prompt for the image generation
+     * Binary image data
      */
-    negative_prompt?: string;
+    data: Blob | ArrayBuffer;
 };
-type TextToImageReturn = Blob;
+interface ImageToTextReturn {
+    /**
+     * The generated caption
+     */
+    generated_text: string;
+}
 declare class HfInference {
     private readonly apiKey;
     private readonly defaultOptions;
@@ -615,6 +645,10 @@ declare class HfInference {
      * Recommended model: stabilityai/stable-diffusion-2
      */
     textToImage(args: TextToImageArgs, options?: Options): Promise<TextToImageReturn>;
+    /**
+     * This task reads some image input and outputs the text caption.
+     */
+    imageToText(args: ImageToTextArgs, options?: Options): Promise<ImageToTextReturn>;
     /**
      * Helper that prepares request arguments
      */
@@ -640,4 +674,4 @@ declare class HfInference {
     }): AsyncGenerator<T>;
 }
 
-export { Args, AudioClassificationArgs, AudioClassificationReturn, AudioClassificationReturnValue, AutomaticSpeechRecognitionArgs, AutomaticSpeechRecognitionReturn, ConversationalArgs, ConversationalReturn, FeatureExtractionArgs, FeatureExtractionReturn, FillMaskArgs, FillMaskReturn, HfInference, ImageClassificationArgs, ImageClassificationReturn, ImageClassificationReturnValue, ImageSegmentationArgs, ImageSegmentationReturn, ImageSegmentationReturnValue, ObjectDetectionArgs, ObjectDetectionReturn, ObjectDetectionReturnValue, Options, QuestionAnswerArgs, QuestionAnswerReturn, SummarizationArgs, SummarizationReturn, TableQuestionAnswerArgs, TableQuestionAnswerReturn, TextClassificationArgs, TextClassificationReturn, TextGenerationArgs, TextGenerationReturn, TextGenerationStreamBestOfSequence, TextGenerationStreamDetails, TextGenerationStreamFinishReason, TextGenerationStreamPrefillToken, TextGenerationStreamReturn, TextGenerationStreamToken, TextToImageArgs, TextToImageReturn, TokenClassificationArgs, TokenClassificationReturn, TokenClassificationReturnValue, TranslationArgs, TranslationReturn, ZeroShotClassificationArgs, ZeroShotClassificationReturn, ZeroShotClassificationReturnValue };
+export { Args, AudioClassificationArgs, AudioClassificationReturn, AudioClassificationReturnValue, AutomaticSpeechRecognitionArgs, AutomaticSpeechRecognitionReturn, ConversationalArgs, ConversationalReturn, FeatureExtractionArgs, FeatureExtractionReturn, FillMaskArgs, FillMaskReturn, HfInference, ImageClassificationArgs, ImageClassificationReturn, ImageClassificationReturnValue, ImageSegmentationArgs, ImageSegmentationReturn, ImageSegmentationReturnValue, ImageToTextArgs, ImageToTextReturn, ObjectDetectionArgs, ObjectDetectionReturn, ObjectDetectionReturnValue, Options, QuestionAnswerArgs, QuestionAnswerReturn, SummarizationArgs, SummarizationReturn, TableQuestionAnswerArgs, TableQuestionAnswerReturn, TextClassificationArgs, TextClassificationReturn, TextGenerationArgs, TextGenerationReturn, TextGenerationStreamBestOfSequence, TextGenerationStreamDetails, TextGenerationStreamFinishReason, TextGenerationStreamPrefillToken, TextGenerationStreamReturn, TextGenerationStreamToken, TextToImageArgs, TextToImageReturn, TokenClassificationArgs, TokenClassificationReturn, TokenClassificationReturnValue, TranslationArgs, TranslationReturn, ZeroShotClassificationArgs, ZeroShotClassificationReturn, ZeroShotClassificationReturnValue };

package/dist/index.js

@@ -1,3 +1,4 @@
+"use strict";
 var __defProp = Object.defineProperty;
 var __getOwnPropDesc = Object.getOwnPropertyDescriptor;
 var __getOwnPropNames = Object.getOwnPropertyNames;
@@ -24,7 +25,7 @@ __export(src_exports, {
 });
 module.exports = __toCommonJS(src_exports);
 
-// src/utils/to-array.ts
+// src/utils/toArray.ts
 function toArray(obj) {
   if (Array.isArray(obj)) {
     return obj;
@@ -386,6 +387,15 @@ var HfInference = class {
     }
     return res;
   }
+  /**
+   * This task reads some image input and outputs the text caption.
+   */
+  async imageToText(args, options) {
+    return (await this.request(args, {
+      ...options,
+      binary: true
+    }))?.[0];
+  }
   /**
    * Helper that prepares request arguments
    */
@@ -465,10 +475,15 @@ var HfInference = class {
       throw new Error(`Server response contains error: ${response.status}`);
     }
     if (response.headers.get("content-type") !== "text/event-stream") {
-      throw new Error(`Server does not support event stream content type`);
+      throw new Error(
+        `Server does not support event stream content type, it returned ` + response.headers.get("content-type")
+      );
+    }
+    if (!response.body) {
+      return;
     }
     const reader = response.body.getReader();
-    const events = [];
+    let events = [];
     const onEvent = (event) => {
       events.push(event);
     };
@@ -487,12 +502,12 @@ var HfInference = class {
         if (done)
           return;
         onChunk(value);
-        while (events.length > 0) {
-          const event = events.shift();
+        for (const event of events) {
           if (event.data.length > 0) {
             yield JSON.parse(event.data);
           }
         }
+        events = [];
       }
     } finally {
       reader.releaseLock();

package/dist/index.mjs

@@ -1,4 +1,4 @@
-// src/utils/to-array.ts
+// src/utils/toArray.ts
 function toArray(obj) {
   if (Array.isArray(obj)) {
     return obj;
@@ -360,6 +360,15 @@ var HfInference = class {
     }
     return res;
   }
+  /**
+   * This task reads some image input and outputs the text caption.
+   */
+  async imageToText(args, options) {
+    return (await this.request(args, {
+      ...options,
+      binary: true
+    }))?.[0];
+  }
   /**
    * Helper that prepares request arguments
    */
@@ -439,10 +448,15 @@ var HfInference = class {
       throw new Error(`Server response contains error: ${response.status}`);
     }
     if (response.headers.get("content-type") !== "text/event-stream") {
-      throw new Error(`Server does not support event stream content type`);
+      throw new Error(
+        `Server does not support event stream content type, it returned ` + response.headers.get("content-type")
+      );
+    }
+    if (!response.body) {
+      return;
     }
     const reader = response.body.getReader();
-    const events = [];
+    let events = [];
     const onEvent = (event) => {
       events.push(event);
     };
@@ -461,12 +475,12 @@ var HfInference = class {
         if (done)
           return;
         onChunk(value);
-        while (events.length > 0) {
-          const event = events.shift();
+        for (const event of events) {
           if (event.data.length > 0) {
             yield JSON.parse(event.data);
           }
         }
+        events = [];
       }
     } finally {
       reader.releaseLock();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@huggingface/inference",
-  "version": "1.7.1",
+  "version": "1.8.0",
   "license": "MIT",
   "author": "Tim Mikeladze <tim.mikeladze@gmail.com>",
   "description": "Typescript wrapper for the Hugging Face Inference API",
@@ -40,10 +40,9 @@
   },
   "devDependencies": {
     "@types/node": "18.13.0",
-    "tsup": "^6.6.3",
     "typescript": "4.9.5",
     "vite": "^4.1.4",
-    "vitest": "^0.29.2"
+    "vitest": "^0.29.8"
   },
   "resolutions": {},
   "scripts": {
@@ -52,8 +51,8 @@
     "lint:check": "eslint --ext .cjs,.ts .",
     "format": "prettier --write .",
     "format:check": "prettier --check .",
-    "test": "vitest run",
-    "test:browser": "vitest run --browser.name=chrome --browser.headless",
+    "test": "vitest run --config vitest.config.ts",
+    "test:browser": "vitest run --browser.name=chrome --browser.headless --config vitest.config.ts",
     "type-check": "tsc"
   }
 }

package/src/HfInference.ts

@@ -1,4 +1,4 @@
-import { toArray } from "./utils/to-array";
+import { toArray } from "./utils/toArray";
 import type { EventSourceMessage } from "./vendor/fetch-event-source/parse";
 import { getLines, getMessages } from "./vendor/fetch-event-source/parse";
 
@@ -299,12 +299,12 @@ export interface TextGenerationStreamReturn {
 	 * Complete generated text
 	 * Only available when the generation is finished
 	 */
-	generated_text?: string;
+	generated_text: string | null;
 	/**
 	 * Generation details
 	 * Only available when the generation is finished
 	 */
-	details?: TextGenerationStreamDetails;
+	details: TextGenerationStreamDetails | null;
 }
 
 export type TokenClassificationArgs = Args & {
@@ -582,13 +582,45 @@ export type TextToImageArgs = Args & {
 	 */
 	inputs: string;
 
+	parameters?: {
+		/**
+		 * An optional negative prompt for the image generation
+		 */
+		negative_prompt?: string;
+		/**
+		 * The height in pixels of the generated image
+		 */
+		height?: number;
+		/**
+		 * The width in pixels of the generated image
+		 */
+		width?: number;
+		/**
+		 * The number of denoising steps. More denoising steps usually lead to a higher quality image at the expense of slower inference.
+		 */
+		num_inference_steps?: number;
+		/**
+		 * Guidance scale: Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, usually at the expense of lower image quality.
+		 */
+		guidance_scale?: number;
+	};
+};
+
+export type TextToImageReturn = Blob;
+
+export type ImageToTextArgs = Args & {
 	/**
-	 * An optional negative prompt for the image generation
+	 * Binary image data
 	 */
-	negative_prompt?: string;
+	data: Blob | ArrayBuffer;
 };
 
-export type TextToImageReturn = Blob;
+export interface ImageToTextReturn {
+	/**
+	 * The generated caption
+	 */
+	generated_text: string;
+}
 
 export class HfInference {
 	private readonly apiKey: string;
@@ -928,6 +960,18 @@ export class HfInference {
 		return res;
 	}
 
+	/**
+	 * This task reads some image input and outputs the text caption.
+	 */
+	public async imageToText(args: ImageToTextArgs, options?: Options): Promise<ImageToTextReturn> {
+		return (
+			await this.request<[ImageToTextReturn]>(args, {
+				...options,
+				binary: true,
+			})
+		)?.[0];
+	}
+
 	/**
 	 * Helper that prepares request arguments
 	 */
@@ -1048,11 +1092,17 @@ export class HfInference {
 			throw new Error(`Server response contains error: ${response.status}`);
 		}
 		if (response.headers.get("content-type") !== "text/event-stream") {
-			throw new Error(`Server does not support event stream content type`);
+			throw new Error(
+				`Server does not support event stream content type, it returned ` + response.headers.get("content-type")
+			);
+		}
+
+		if (!response.body) {
+			return;
 		}
 
 		const reader = response.body.getReader();
-		const events: EventSourceMessage[] = [];
+		let events: EventSourceMessage[] = [];
 
 		const onEvent = (event: EventSourceMessage) => {
 			// accumulate events in array
@@ -1072,12 +1122,12 @@ export class HfInference {
 				const { done, value } = await reader.read();
 				if (done) return;
 				onChunk(value);
-				while (events.length > 0) {
-					const event = events.shift();
+				for (const event of events) {
 					if (event.data.length > 0) {
 						yield JSON.parse(event.data) as T;
 					}
 				}
+				events = [];
 			}
 		} finally {
 			reader.releaseLock();

package/src/utils/omit.ts

@@ -0,0 +1,11 @@
+import { pick } from "./pick";
+import { typedInclude } from "./typedInclude";
+
+/**
+ * Return copy of object, omitting blocklisted array of props
+ */
+export function omit<T extends object, K extends keyof T>(o: T, props: K[] | K): Pick<T, Exclude<keyof T, K>> {
+	const propsArr = Array.isArray(props) ? props : [props];
+	const letsKeep = (Object.keys(o) as (keyof T)[]).filter((prop) => !typedInclude(propsArr, prop));
+	return pick(o, letsKeep);
+}

package/src/utils/pick.ts

@@ -0,0 +1,16 @@
+/**
+ * Return copy of object, only keeping allowlisted properties.
+ *
+ * This doesn't add {p: undefined} anymore, for props not in the o object.
+ */
+export function pick<T, K extends keyof T>(o: T, props: K[] | ReadonlyArray<K>): Pick<T, K> {
+	// inspired by stackoverflow.com/questions/25553910/one-liner-to-take-some-properties-from-object-in-es-6
+	return Object.assign(
+		{},
+		...props.map((prop) => {
+			if (o[prop] !== undefined) {
+				return { [prop]: o[prop] };
+			}
+		})
+	);
+}

package/src/utils/typedInclude.ts

@@ -0,0 +1,3 @@
+export function typedInclude<V, T extends V>(arr: readonly T[], v: V): v is T {
+	return arr.includes(v as T);
+}

package/src/utils/env-predicates.ts

@@ -1,7 +0,0 @@
-const isBrowser = typeof window !== "undefined" && typeof window.document !== "undefined";
-
-const isWebWorker =
-	typeof self === "object" && self.constructor && self.constructor.name === "DedicatedWorkerGlobalScope";
-
-export const isFrontend = isBrowser || isWebWorker;
-export const isBackend = !isBrowser && !isWebWorker;