@huggingface/transformers 3.0.1 → 3.1.0

package/src/pipelines.js

@@ -45,8 +45,10 @@ import {
 } from './models.js';
 import {
     AutoProcessor,
-    Processor
-} from './processors.js';
+} from './models/auto/processing_auto.js';
+import {
+    Processor,
+} from './base/processing_utils.js';
 
 import {
     Callable,
@@ -54,7 +56,6 @@ import {
 
 import {
     dispatchCallback,
-    pop,
     product,
 } from './utils/core.js';
 import {
@@ -158,7 +159,6 @@ function get_bounding_box(box, asInteger) {
 /**
  * The Pipeline class is the class from which all pipelines inherit.
  * Refer to this class for methods shared across different pipelines.
- * @extends Callable
  */
 export class Pipeline extends Callable {
     /**
@@ -2131,8 +2131,8 @@ export class ImageSegmentationPipeline extends (/** @type {new (options: ImagePi
             fn = this.subtasks_mapping[subtask];
         } else {
             for (let [task, func] of Object.entries(this.subtasks_mapping)) {
-                if (func in this.processor.feature_extractor) {
-                    fn = this.processor.feature_extractor[func].bind(this.processor.feature_extractor);
+                if (func in this.processor.image_processor) {
+                    fn = this.processor.image_processor[func].bind(this.processor.image_processor);
                     subtask = task;
                     break;
                 }
@@ -2362,7 +2362,7 @@ export class ObjectDetectionPipeline extends (/** @type {new (options: ImagePipe
         const output = await this.model({ pixel_values, pixel_mask });
 
         // @ts-ignore
-        const processed = this.processor.feature_extractor.post_process_object_detection(output, threshold, imageSizes);
+        const processed = this.processor.image_processor.post_process_object_detection(output, threshold, imageSizes);
 
         // Add labels
         const id2label = this.model.config.id2label;
@@ -2510,7 +2510,7 @@ export class ZeroShotObjectDetectionPipeline extends (/** @type {new (options: T
             const output = await this.model({ ...text_inputs, pixel_values });
 
             // @ts-ignore
-            const processed = this.processor.feature_extractor.post_process_object_detection(output, threshold, imageSize, true)[0];
+            const processed = this.processor.image_processor.post_process_object_detection(output, threshold, imageSize, true)[0];
             let result = processed.boxes.map((box, i) => ({
                 score: processed.scores[i],
                 label: candidate_labels[processed.classes[i]],

package/src/tokenizers.js

@@ -1518,6 +1518,8 @@ class SplitPreTokenizer extends PreTokenizer {
 
         if (this.config.invert) {
             return text.match(this.pattern) || [];
+        } else if (this.config.behavior?.toLowerCase() === 'removed') {
+            return text.split(this.pattern).filter(x => x);
         } else {
             return regexSplit(text, this.pattern);
         }
@@ -4255,6 +4257,8 @@ export class VitsTokenizer extends PreTrainedTokenizer {
 
 export class CohereTokenizer extends PreTrainedTokenizer { }
 
+export class MgpstrTokenizer extends PreTrainedTokenizer { }
+
 /**
  * Helper class which is used to instantiate pretrained tokenizers with the `from_pretrained` function.
  * The chosen tokenizer class is determined by the type specified in the tokenizer config.
@@ -4308,6 +4312,7 @@ export class AutoTokenizer {
         GemmaTokenizer,
         Grok1Tokenizer,
         CohereTokenizer,
+        MgpstrTokenizer,
 
         // Base case:
         PreTrainedTokenizer,

package/src/transformers.js

@@ -12,10 +12,10 @@
  */
 
 export { env } from './env.js';
+
 export * from './pipelines.js';
 export * from './models.js';
 export * from './tokenizers.js';
-export * from './processors.js';
 export * from './configs.js';
 
 export * from './utils/audio.js';
@@ -23,6 +23,19 @@ export * from './utils/image.js';
 export * from './utils/tensor.js';
 export * from './utils/maths.js';
 
+
+export { FeatureExtractor } from './base/feature_extraction_utils.js';
+export * from './models/feature_extractors.js';
+export * from './models/auto/feature_extraction_auto.js';
+
+export { ImageProcessor } from './base/image_processors_utils.js';
+export * from './models/image_processors.js';
+export * from './models/auto/image_processing_auto.js';
+
+export { Processor } from './base/processing_utils.js';
+export * from './models/processors.js';
+export * from './models/auto/processing_auto.js';
+
 export * from './generation/streamers.js';
 export * from './generation/stopping_criteria.js';
-
+export * from './generation/logits_process.js';

package/src/utils/constants.js

@@ -1,2 +1,9 @@
 
-export const GITHUB_ISSUE_URL = 'https://github.com/huggingface/transformers.js/issues/new/choose';
+export const GITHUB_ISSUE_URL = 'https://github.com/huggingface/transformers.js/issues/new/choose';
+
+export const CONFIG_NAME = "config.json"
+export const FEATURE_EXTRACTOR_NAME = "preprocessor_config.json"
+export const IMAGE_PROCESSOR_NAME = FEATURE_EXTRACTOR_NAME
+export const PROCESSOR_NAME = "processor_config.json"
+export const CHAT_TEMPLATE_NAME = "chat_template.json"
+export const GENERATION_CONFIG_NAME = "generation_config.json"

package/src/utils/core.js

@@ -1,18 +1,37 @@
 
 /**
  * @file Core utility functions/classes for Transformers.js.
- * 
+ *
  * These are only used internally, meaning an end-user shouldn't
  * need to access anything here.
- * 
+ *
  * @module utils/core
  */
 
+/**
+ * @typedef {Object} ProgressInfo
+ * @property {'initiate' | 'download' | 'progress' | 'done'} status The status of the progress item.
+ * @property {string} name This can be either:
+ * - a string, the *model id* of a model repo on huggingface.co.
+ * - a path to a *directory* potentially containing the file.
+ * @property {string} file The name of the file
+ * @property {number} [progress] A number between 0 and 100. Only available for the 'progress' status.
+ * @property {number} [loaded] The number of bytes loaded. Only available for the 'progress' status.
+ * @property {number} [total] The total number of bytes to be loaded. Only available for the 'progress' status.
+ */
+
+/**
+ * A callback function that is called with progress information.
+ * @callback ProgressCallback
+ * @param {ProgressInfo} progressInfo
+ * @returns {void}
+ */
+
 /**
  * Helper function to dispatch progress callbacks.
  *
- * @param {Function} progress_callback The progress callback function to dispatch.
- * @param {any} data The data to pass to the progress callback function.
+ * @param {ProgressCallback | null | undefined} progress_callback The progress callback function to dispatch.
+ * @param {ProgressInfo} data The data to pass to the progress callback function.
  * @returns {void}
  * @private
  */
@@ -46,7 +65,7 @@ export function escapeRegExp(string) {
  * Check if a value is a typed array.
  * @param {*} val The value to check.
  * @returns {boolean} True if the value is a `TypedArray`, false otherwise.
- * 
+ *
  * Adapted from https://stackoverflow.com/a/71091338/13989043
  */
 export function isTypedArray(val) {
@@ -63,6 +82,15 @@ export function isIntegralNumber(x) {
     return Number.isInteger(x) || typeof x === 'bigint'
 }
 
+/**
+ * Determine if a provided width or height is nullish.
+ * @param {*} x The value to check.
+ * @returns {boolean} True if the value is `null`, `undefined` or `-1`, false otherwise.
+ */
+export function isNullishDimension(x) {
+    return x === null || x === undefined || x === -1;
+}
+
 /**
  * Calculates the dimensions of a nested array.
  *
@@ -132,9 +160,9 @@ export function calculateReflectOffset(i, w) {
 }
 
 /**
- * 
- * @param {Object} o 
- * @param {string[]} props 
+ *
+ * @param {Object} o
+ * @param {string[]} props
  * @returns {Object}
  */
 export function pick(o, props) {
@@ -151,7 +179,7 @@ export function pick(o, props) {
 /**
  * Calculate the length of a string, taking multi-byte characters into account.
  * This mimics the behavior of Python's `len` function.
- * @param {string} s The string to calculate the length of. 
+ * @param {string} s The string to calculate the length of.
  * @returns {number} The length of the string.
  */
 export function len(s) {

package/src/utils/hub.js

@@ -13,7 +13,7 @@ import { dispatchCallback } from './core.js';
 
 /**
  * @typedef {Object} PretrainedOptions Options for loading a pretrained model.     
- * @property {function} [progress_callback=null] If specified, this function will be called during model construction, to provide the user with progress updates.
+ * @property {import('./core.js').ProgressCallback} [progress_callback=null] If specified, this function will be called during model construction, to provide the user with progress updates.
  * @property {import('../configs.js').PretrainedConfig} [config=null] Configuration for the model to use instead of an automatically loaded configuration. Configuration can be automatically loaded when:
  * - The model is a model provided by the library (loaded with the *model id* string of a pretrained model).
  * - The model is loaded by supplying a local directory as `pretrained_model_name_or_path` and a configuration JSON file named *config.json* is found in the directory.
@@ -504,6 +504,7 @@ export async function getModelFile(path_or_repo_id, filename, fatal = true, opti
         file: filename
     })
 
+    /** @type {import('./core.js').ProgressInfo} */
     const progressInfo = {
         status: 'progress',
         name: path_or_repo_id,

package/src/utils/image.js

@@ -1,13 +1,14 @@
 
 /**
- * @file Helper module for image processing. 
- * 
- * These functions and classes are only used internally, 
+ * @file Helper module for image processing.
+ *
+ * These functions and classes are only used internally,
  * meaning an end-user shouldn't need to access anything here.
- * 
+ *
  * @module utils/image
  */
 
+import { isNullishDimension } from './core.js';
 import { getFile } from './hub.js';
 import { env } from '../env.js';
 import { Tensor } from './tensor.js';
@@ -91,7 +92,7 @@ export class RawImage {
         this.channels = channels;
     }
 
-    /** 
+    /**
      * Returns the size of the image (width, height).
      * @returns {[number, number]} The size of the image (width, height).
      */
@@ -101,9 +102,9 @@ export class RawImage {
 
     /**
      * Helper method for reading an image from a variety of input types.
-     * @param {RawImage|string|URL} input 
+     * @param {RawImage|string|URL} input
      * @returns The image object.
-     * 
+     *
      * **Example:** Read image from a URL.
      * ```javascript
      * let image = await RawImage.read('https://huggingface.co/datasets/Xenova/transformers.js-docs/resolve/main/football-match.jpg');
@@ -181,7 +182,7 @@ export class RawImage {
 
     /**
      * Helper method to create a new Image from a tensor
-     * @param {Tensor} tensor 
+     * @param {Tensor} tensor
      */
     static fromTensor(tensor, channel_format = 'CHW') {
         if (tensor.dims.length !== 3) {
@@ -306,8 +307,8 @@ export class RawImage {
 
     /**
      * Resize the image to the given dimensions. This method uses the canvas API to perform the resizing.
-     * @param {number} width The width of the new image.
-     * @param {number} height The height of the new image.
+     * @param {number} width The width of the new image. `null` or `-1` will preserve the aspect ratio.
+     * @param {number} height The height of the new image. `null` or `-1` will preserve the aspect ratio.
      * @param {Object} options Additional options for resizing.
      * @param {0|1|2|3|4|5|string} [options.resample] The resampling method to use.
      * @returns {Promise<RawImage>} `this` to support chaining.
@@ -316,9 +317,28 @@ export class RawImage {
         resample = 2,
     } = {}) {
 
+        // Do nothing if the image already has the desired size
+        if (this.width === width && this.height === height) {
+            return this;
+        }
+
         // Ensure resample method is a string
         let resampleMethod = RESAMPLING_MAPPING[resample] ?? resample;
 
+        // Calculate width / height to maintain aspect ratio, in the event that
+        // the user passed a null value in.
+        // This allows users to pass in something like `resize(320, null)` to
+        // resize to 320 width, but maintain aspect ratio.
+        const nullish_width = isNullishDimension(width);
+        const nullish_height = isNullishDimension(height);
+        if (nullish_width && nullish_height) {
+            return this;
+        } else if (nullish_width) {
+            width = (height / this.height) * this.width;
+        } else if (nullish_height) {
+            height = (width / this.width) * this.height;
+        }
+
         if (BROWSER_ENV) {
             // TODO use `resample` in browser environment
 
@@ -355,7 +375,7 @@ export class RawImage {
                 case 'nearest':
                 case 'bilinear':
                 case 'bicubic':
-                    // Perform resizing using affine transform. 
+                    // Perform resizing using affine transform.
                     // This matches how the python Pillow library does it.
                     img = img.affine([width / this.width, 0, 0, height / this.height], {
                         interpolator: resampleMethod
@@ -368,7 +388,7 @@ export class RawImage {
                     img = img.resize({
                         width, height,
                         fit: 'fill',
-                        kernel: 'lanczos3', // PIL Lanczos uses a kernel size of 3 
+                        kernel: 'lanczos3', // PIL Lanczos uses a kernel size of 3
                     });
                     break;
 
@@ -408,13 +428,14 @@ export class RawImage {
             // Draw image to context, padding in the process
             ctx.drawImage(canvas,
                 0, 0, this.width, this.height,
-                left, top, newWidth, newHeight
+                left, top, this.width, this.height
             );
 
             // Create image from the padded data
             const paddedImage = new RawImage(
                 ctx.getImageData(0, 0, newWidth, newHeight).data,
-                newWidth, newHeight, 4);
+                newWidth, newHeight, 4
+            );
 
             // Convert back so that image has the same number of channels as before
             return paddedImage.convert(numChannels);
@@ -447,7 +468,7 @@ export class RawImage {
             // Create canvas object for this image
             const canvas = this.toCanvas();
 
-            // Create a new canvas of the desired size. This is needed since if the 
+            // Create a new canvas of the desired size. This is needed since if the
             // image is too small, we need to pad it with black pixels.
             const ctx = createCanvasFunction(crop_width, crop_height).getContext('2d');
 
@@ -495,7 +516,7 @@ export class RawImage {
             // Create canvas object for this image
             const canvas = this.toCanvas();
 
-            // Create a new canvas of the desired size. This is needed since if the 
+            // Create a new canvas of the desired size. This is needed since if the
             // image is too small, we need to pad it with black pixels.
             const ctx = createCanvasFunction(crop_width, crop_height).getContext('2d');
 
@@ -637,6 +658,36 @@ export class RawImage {
         return clonedCanvas;
     }
 
+    /**
+     * Split this image into individual bands. This method returns an array of individual image bands from an image.
+     * For example, splitting an "RGB" image creates three new images each containing a copy of one of the original bands (red, green, blue).
+     * 
+     * Inspired by PIL's `Image.split()` [function](https://pillow.readthedocs.io/en/latest/reference/Image.html#PIL.Image.Image.split).
+     * @returns {RawImage[]} An array containing bands.
+     */
+    split() {
+        const { data, width, height, channels } = this;
+
+        /** @type {typeof Uint8Array | typeof Uint8ClampedArray} */
+        const data_type = /** @type {any} */(data.constructor);
+        const per_channel_length = data.length / channels;
+
+        // Pre-allocate buffers for each channel
+        const split_data = Array.from(
+            { length: channels },
+            () => new data_type(per_channel_length),
+        );
+
+        // Write pixel data
+        for (let i = 0; i < per_channel_length; ++i) {
+            const data_offset = channels * i;
+            for (let j = 0; j < channels; ++j) {
+                split_data[j][i] = data[data_offset + j];
+            }
+        }
+        return split_data.map((data) => new RawImage(data, width, height, 1));
+    }
+
     /**
      * Helper method to update the image data.
      * @param {Uint8ClampedArray} data The new image data.
@@ -742,4 +793,4 @@ export class RawImage {
             }
         });
     }
-}
+}

package/src/utils/tensor.js

@@ -340,10 +340,43 @@ export class Tensor {
         return this;
     }
 
+    /**
+     * Creates a deep copy of the current Tensor.
+     * @returns {Tensor} A new Tensor with the same type, data, and dimensions as the original.
+     */
     clone() {
         return new Tensor(this.type, this.data.slice(), this.dims.slice());
     }
 
+    /**
+     * Performs a slice operation on the Tensor along specified dimensions.
+     *
+     * Consider a Tensor that has a dimension of [4, 7]:
+     * ```
+     * [ 1,  2,  3,  4,  5,  6,  7]
+     * [ 8,  9, 10, 11, 12, 13, 14]
+     * [15, 16, 17, 18, 19, 20, 21]
+     * [22, 23, 24, 25, 26, 27, 28]
+     * ```
+     * We can slice against the two dims of row and column, for instance in this
+     * case we can start at the second element, and return to the second last,
+     * like this:
+     * ```
+     * tensor.slice([1, -1], [1, -1]);
+     * ```
+     * which would return:
+     * ```
+     * [  9, 10, 11, 12, 13 ]
+     * [ 16, 17, 18, 19, 20 ]
+     * ```
+     *
+     * @param {...(number|number[]|null)} slices The slice specifications for each dimension.
+     * - If a number is given, then a single element is selected.
+     * - If an array of two numbers is given, then a range of elements [start, end (exclusive)] is selected.
+     * - If null is given, then the entire dimension is selected.
+     * @returns {Tensor} A new Tensor containing the selected elements.
+     * @throws {Error} If the slice input is invalid.
+     */
     slice(...slices) {
         // This allows for slicing with ranges and numbers
         const newTensorDims = [];
@@ -413,7 +446,6 @@ export class Tensor {
             data[i] = this_data[originalIndex];
         }
         return new Tensor(this.type, data, newTensorDims);
-
     }
 
     /**

package/types/base/feature_extraction_utils.d.ts

@@ -0,0 +1,41 @@
+/**
+ * Helper function to validate audio inputs.
+ * @param {any} audio The audio data.
+ * @param {string} feature_extractor The name of the feature extractor.
+ * @private
+ */
+export function validate_audio_inputs(audio: any, feature_extractor: string): void;
+declare const FeatureExtractor_base: new () => {
+    (...args: any[]): any;
+    _call(...args: any[]): any;
+};
+/**
+ * Base class for feature extractors.
+ */
+export class FeatureExtractor extends FeatureExtractor_base {
+    /**
+     * Instantiate one of the processor classes of the library from a pretrained model.
+     *
+     * The processor class to instantiate is selected based on the `image_processor_type` (or `feature_extractor_type`; legacy)
+     * property of the config object (either passed as an argument or loaded from `pretrained_model_name_or_path` if possible)
+     *
+     * @param {string} pretrained_model_name_or_path The name or path of the pretrained model. Can be either:
+     * - A string, the *model id* of a pretrained processor hosted inside a model repo on huggingface.co.
+     *   Valid model ids can be located at the root-level, like `bert-base-uncased`, or namespaced under a
+     *   user or organization name, like `dbmdz/bert-base-german-cased`.
+     * - A path to a *directory* containing processor files, e.g., `./my_model_directory/`.
+     * @param {import('../utils/hub.js').PretrainedOptions} options Additional options for loading the processor.
+     *
+     * @returns {Promise<FeatureExtractor>} A new instance of the Processor class.
+     */
+    static from_pretrained(pretrained_model_name_or_path: string, options: import('../utils/hub.js').PretrainedOptions): Promise<FeatureExtractor>;
+    /**
+     * Constructs a new FeatureExtractor instance.
+     *
+     * @param {Object} config The configuration for the feature extractor.
+     */
+    constructor(config: any);
+    config: any;
+}
+export {};
+//# sourceMappingURL=feature_extraction_utils.d.ts.map

package/types/base/feature_extraction_utils.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"feature_extraction_utils.d.ts","sourceRoot":"","sources":["../../src/base/feature_extraction_utils.js"],"names":[],"mappings":"AAwCA;;;;;GAKG;AACH,6CAJW,GAAG,qBACH,MAAM,QAUhB;;;;;AAjDD;;GAEG;AACH;IAWI;;;;;;;;;;;;;;OAcG;IACH,sDATW,MAAM,WAKN,OAAO,iBAAiB,EAAE,iBAAiB,GAEzC,QAAQ,gBAAgB,CAAC,CAKrC;IA5BD;;;;OAIG;IACH,yBAGC;IADG,YAAoB;CAsB3B"}