@huggingface/transformers - npm Package Compare versions

Comparing version 3.2.3 to 3.2.4

package.json

 {
   "name": "@huggingface/transformers",
-  "version": "3.2.3",
+  "version": "3.2.4",
   "description": "State-of-the-art Machine Learning for the web. Run 🤗 Transformers directly in your browser, with no need for a server!",
@@ -27,3 +27,3 @@ "main": "./src/transformers.js",
     "format:check": "prettier --check .",
-    "typegen": "tsc ./src/transformers.js --allowJs --declaration --emitDeclarationOnly --declarationMap --outDir types",
+    "typegen": "tsc --build",
     "dev": "webpack serve --no-client-overlay",
@@ -30,0 +30,0 @@ "build": "webpack && npm run typegen",

src/base/feature_extraction_utils.js

@@ -20,19 +20,19 @@ import { FEATURE_EXTRACTOR_NAME } from "../utils/constants.js";
     /**
-     * Instantiate one of the processor classes of the library from a pretrained model.
+     * Instantiate one of the feature extractor 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)
+     * The feature extractor class to instantiate is selected based on the `feature_extractor_type` 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.
+     * - A string, the *model id* of a pretrained feature_extractor 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.
+     * - A path to a *directory* containing feature_extractor files, e.g., `./my_model_directory/`.
+     * @param {import('../utils/hub.js').PretrainedOptions} options Additional options for loading the feature_extractor.
      * 
-     * @returns {Promise<FeatureExtractor>} A new instance of the Processor class.
+     * @returns {Promise<FeatureExtractor>} A new instance of the Feature Extractor class.
      */
     static async from_pretrained(pretrained_model_name_or_path, options) {
-        const preprocessorConfig = await getModelJSON(pretrained_model_name_or_path, FEATURE_EXTRACTOR_NAME, true, options);
-        return new this(preprocessorConfig);
+        const config = await getModelJSON(pretrained_model_name_or_path, FEATURE_EXTRACTOR_NAME, true, options);
+        return new this(config);
     }
@@ -39,0 +39,0 @@ }

src/base/image_processors_utils.js

@@ -607,10 +607,16 @@ import { Callable } from "../utils/generic.js";
         this.do_resize = config.do_resize ?? (this.size !== undefined);
+        // @ts-expect-error TS2339
         this.size_divisibility = config.size_divisibility ?? config.size_divisor;
 
         this.do_center_crop = config.do_center_crop;
+        // @ts-expect-error TS2339
         this.crop_size = config.crop_size;
+        // @ts-expect-error TS2339
         this.do_convert_rgb = config.do_convert_rgb ?? true;
+        // @ts-expect-error TS2339
         this.do_crop_margin = config.do_crop_margin;
 
+        // @ts-expect-error TS2339
         this.pad_size = config.pad_size;
+        // @ts-expect-error TS2339
         this.do_pad = config.do_pad;
@@ -824,2 +830,3 @@
             shortest_edge = size;
+            // @ts-expect-error TS2339
             longest_edge = this.config.max_size ?? shortest_edge;
@@ -893,2 +900,3 @@
             const { min_pixels, max_pixels } = size;
+            // @ts-expect-error TS2339
             const factor = this.config.patch_size * this.config.merge_size;
@@ -909,2 +917,3 @@ return smart_resize(srcHeight, srcWidth, factor, min_pixels, max_pixels);
         return await image.resize(newWidth, newHeight, {
+            // @ts-expect-error TS2322
             resample: this.resample,
@@ -960,2 +969,3 @@ });
         if (this.do_thumbnail) {
+            // @ts-expect-error TS2345
             image = await this.thumbnail(image, this.size, this.resample);
@@ -985,2 +995,3 @@ }
         // to emulate the behavior of the original Python code (w/ numpy).
+        /** @type {Float32Array} */
         let pixelData = Float32Array.from(image.data);
@@ -987,0 +998,0 @@ let imgDims = [image.height, image.width, image.channels];

src/base/processing_utils.js

@@ -31,2 +31,3 @@
  * @typedef {import('../utils/hub.js').PretrainedOptions & ProcessorProperties} PretrainedProcessorOptions
+ * @typedef {import('../tokenizers.js').PreTrainedTokenizer} PreTrainedTokenizer
  */
@@ -65,3 +66,3 @@
     /**
-     * @returns {import('../tokenizers.js').PreTrainedTokenizer|undefined} The tokenizer of the processor, if it exists.
+     * @returns {PreTrainedTokenizer|undefined} The tokenizer of the processor, if it exists.
      */
@@ -79,2 +80,7 @@ get tokenizer() {
 
+    /**
+     * @param {Parameters<PreTrainedTokenizer['apply_chat_template']>[0]} messages
+     * @param {Parameters<PreTrainedTokenizer['apply_chat_template']>[1]} options
+     * @returns {ReturnType<PreTrainedTokenizer['apply_chat_template']>}
+     */
     apply_chat_template(messages, options = {}) {
@@ -90,2 +96,6 @@ if (!this.tokenizer) {
 
+    /**
+     * @param {Parameters<PreTrainedTokenizer['batch_decode']>} args
+     * @returns {ReturnType<PreTrainedTokenizer['batch_decode']>}
+     */
     batch_decode(...args) {
@@ -118,4 +128,4 @@ if (!this.tokenizer) {
      * 
-     * The processor class to instantiate is selected based on the `feature_extractor_type` property of the config object
-     * (either passed as an argument or loaded from `pretrained_model_name_or_path` if possible)
+     * 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)
      * 
@@ -122,0 +132,0 @@ * @param {string} pretrained_model_name_or_path The name or path of the pretrained model. Can be either:

src/configs.js

@@ -73,11 +73,15 @@
         case 'idefics3':
+            // @ts-expect-error TS2339
             init_normalized_config = getNormalizedConfig(config.text_config);
             break;
         case 'moondream1':
+            // @ts-expect-error TS2339
             init_normalized_config = getNormalizedConfig(config.phi_config);
             break;
         case 'musicgen':
+            // @ts-expect-error TS2339
             init_normalized_config = getNormalizedConfig(config.decoder);
             break;
         case 'multi_modality':
+            // @ts-expect-error TS2339
             init_normalized_config = getNormalizedConfig(config.language_config);
@@ -203,2 +207,3 @@ break;
         case 'vision-encoder-decoder':
+            // @ts-expect-error TS2339
             const decoderConfig = getNormalizedConfig(config.decoder);
@@ -205,0 +210,0 @@

src/env.js

@@ -29,3 +29,3 @@ /**
 
-const VERSION = '3.2.3';
+const VERSION = '3.2.4';
 
@@ -32,0 +32,0 @@ // Check if various APIs are available (depends on environment)

src/models/auto/feature_extraction_auto.js

@@ -9,18 +9,2 @@
 
-    /**
-     * Instantiate one of the feature extractor classes of the library from a pretrained model.
-     * 
-     * The processor class to instantiate is selected based on the `feature_extractor_type` 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<AllFeatureExtractors.ImageProcessor>} A new instance of the Processor class.
-     */
-
     /** @type {typeof FeatureExtractor.from_pretrained} */
@@ -27,0 +11,0 @@ static async from_pretrained(pretrained_model_name_or_path, options={}) {

src/models/auto/processing_auto.js

@@ -43,18 +43,2 @@
 
-    /**
-     * 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<Processor>} A new instance of the Processor class.
-     */
-
     /** @type {typeof Processor.from_pretrained} */
@@ -61,0 +45,0 @@ static async from_pretrained(pretrained_model_name_or_path, options={}) {

src/models/convnext/image_processing_convnext.js

@@ -12,2 +12,3 @@ import {
          */
+        // @ts-expect-error TS2339
         this.crop_pct = this.config.crop_pct ?? (224 / 256);
@@ -14,0 +15,0 @@ }

src/models/efficientnet/image_processing_efficientnet.js

@@ -8,2 +8,3 @@ import {
         super(config);
+        // @ts-expect-error TS2339
         this.include_top = this.config.include_top ?? true;
@@ -10,0 +11,0 @@ if (this.include_top) {

src/models/florence2/processing_florence2.js

@@ -13,4 +13,7 @@ import { Processor } from "../../base/processing_utils.js";
         const {
+            // @ts-expect-error TS2339
             tasks_answer_post_processing_type,
+            // @ts-expect-error TS2339
             task_prompts_without_inputs,
+            // @ts-expect-error TS2339
             task_prompts_with_input,
@@ -17,0 +20,0 @@ } = this.image_processor.config;

src/models/idefics3/image_processing_idefics3.js

@@ -149,2 +149,4 @@
                     const end_offset = (i + 1) * pixel_attention_mask_stride;
+
+                    // @ts-expect-error
                     pixel_attention_mask_data.fill(false, start_offset, end_offset);
@@ -151,0 +153,0 @@ }

src/models/janus/image_processing_janus.js

@@ -16,2 +16,3 @@
         });
+        // @ts-expect-error TS2339
         this.constant_values = this.config.background_color.map(x => x * this.rescale_factor)
@@ -18,0 +19,0 @@ }

src/models/mgp_str/processing_mgp_str.js

@@ -122,2 +122,4 @@ import { Processor } from "../../base/processing_utils.js";
      */
+    // @ts-expect-error The type of this method is not compatible with the one
+    // in the base class. It might be a good idea to fix this.
     batch_decode([char_logits, bpe_logits, wp_logits]) {
@@ -124,0 +126,0 @@ const [char_preds, char_scores] = this._decode_helper(char_logits, 'char');

src/models/paligemma/processing_paligemma.js

@@ -44,2 +44,3 @@ import { Processor } from "../../base/processing_utils.js";
         const bos_token = this.tokenizer.bos_token;
+        // @ts-expect-error TS2339
         const image_seq_length = this.image_processor.config.image_seq_length;
@@ -46,0 +47,0 @@ let input_strings;

src/models/phi3_v/processing_phi3_v.js

@@ -17,3 +17,3 @@ import { Processor } from "../../base/processing_utils.js";
      * @param {RawImage|RawImage[]} images 
-     * @param  {...any} args 
+     * @param  { { padding?: boolean, truncation?: boolean, num_crops?: number } | undefined } options
      * @returns {Promise<any>}
@@ -20,0 +20,0 @@ */

src/models/pyannote/feature_extraction_pyannote.js

@@ -55,2 +55,3 @@ import { FeatureExtractor, validate_audio_inputs } from '../../base/feature_extraction_utils.js';
             for (let i = 0; i < scores.length; ++i) {
+                /** @type {number[]} */
                 const probabilities = softmax(scores[i]);
@@ -57,0 +58,0 @@ const [score, id] = max(probabilities);

src/models/qwen2_vl/processing_qwen2_vl.js

@@ -31,2 +31,3 @@ import { Processor } from "../../base/processing_utils.js";
         if (image_grid_thw) {
+            // @ts-expect-error TS2551
             let merge_length = this.image_processor.config.merge_size ** 2;
@@ -33,0 +34,0 @@ let index = 0;

src/models/seamless_m4t/feature_extraction_seamless_m4t.js

@@ -136,4 +136,4 @@ import { FeatureExtractor, validate_audio_inputs } from '../../base/feature_extraction_utils.js';
                         [1, numPaddedFrames],
-                    )
-                    padded_attention_mask.data.fill(1n, 0, num_frames);
+                    );
+                    /** @type {BigInt64Array} */ (padded_attention_mask.data).fill(1n, 0, num_frames);
                 }
@@ -140,0 +140,0 @@ }

src/models/whisper/feature_extraction_whisper.js

@@ -47,3 +47,3 @@ import { FeatureExtractor, validate_audio_inputs } from '../../base/feature_extraction_utils.js';
         const data = features.data;
-        const maxValue = max(data)[0];
+        const maxValue = max(/** @type {Float32Array} */(data))[0];
 
@@ -50,0 +50,0 @@ for (let i = 0; i < data.length; ++i) {

src/ops/registry.js

@@ -39,2 +39,12 @@ import { createInferenceSession, isONNXProxy } from "../backends/onnx.js";
 
+    static get nearest_interpolate_4d() {
+        if (!this._nearest_interpolate_4d) {
+            this._nearest_interpolate_4d = wrap(
+                [8, 10, 18, 0, 58, 129, 1, 10, 41, 10, 1, 120, 10, 0, 10, 0, 10, 1, 115, 18, 1, 121, 34, 6, 82, 101, 115, 105, 122, 101, 42, 18, 10, 4, 109, 111, 100, 101, 34, 7, 110, 101, 97, 114, 101, 115, 116, 160, 1, 3, 18, 1, 114, 90, 31, 10, 1, 120, 18, 26, 10, 24, 8, 1, 18, 20, 10, 3, 18, 1, 98, 10, 3, 18, 1, 99, 10, 3, 18, 1, 104, 10, 3, 18, 1, 119, 90, 15, 10, 1, 115, 18, 10, 10, 8, 8, 7, 18, 4, 10, 2, 8, 4, 98, 31, 10, 1, 121, 18, 26, 10, 24, 8, 1, 18, 20, 10, 3, 18, 1, 98, 10, 3, 18, 1, 99, 10, 3, 18, 1, 104, 10, 3, 18, 1, 119, 66, 2, 16, 21],
+                this.session_options,
+                'y',
+            );
+        }
+        return this._nearest_interpolate_4d;
+    }
     static get bilinear_interpolate_4d() {
@@ -41,0 +51,0 @@ if (!this._bilinear_interpolate_4d) {

src/utils/dtypes.js

@@ -0,1 +1,3 @@
+/// <reference types="@webgpu/types" />
+
 import { apis } from "../env.js";
@@ -2,0 +4,0 @@

src/utils/hub.js

@@ -124,3 +124,3 @@
         const data = await fs.promises.readFile(this.filePath);
-        return data.buffer;
+        return /** @type {ArrayBuffer} */ (data.buffer);
     }
@@ -127,0 +127,0 @@

src/utils/maths.js

@@ -228,4 +228,5 @@
  * Returns the value and index of the minimum element in an array.
- * @param {number[]|TypedArray} arr array of numbers.
- * @returns {[number, number]} the value and index of the minimum element, of the form: [valueOfMin, indexOfMin]
+ * @template {number[]|bigint[]|AnyTypedArray} T
+ * @param {T} arr array of numbers.
+ * @returns {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} the value and index of the minimum element, of the form: [valueOfMin, indexOfMin]
  * @throws {Error} If array is empty.
@@ -243,3 +244,3 @@ */
     }
-    return [min, indexOfMin];
+    return /** @type {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} */([min, indexOfMin]);
 }
@@ -250,4 +251,5 @@
  * Returns the value and index of the maximum element in an array.
- * @param {number[]|AnyTypedArray} arr array of numbers.
- * @returns {[number, number]} the value and index of the maximum element, of the form: [valueOfMax, indexOfMax]
+ * @template {number[]|bigint[]|AnyTypedArray} T
+ * @param {T} arr array of numbers.
+ * @returns {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} the value and index of the maximum element, of the form: [valueOfMax, indexOfMax]
  * @throws {Error} If array is empty.
@@ -265,3 +267,3 @@ */
     }
-    return [Number(max), indexOfMax];
+    return /** @type {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} */([max, indexOfMax]);
 }
@@ -268,0 +270,0 @@

src/utils/tensor.js

@@ -12,2 +12,4 @@ /**
     interpolate_data,
+    max,
+    min,
     permute_data
@@ -468,4 +470,2 @@ } from './maths.js';
 
-    // TODO add .max() and .min() methods
-
     /**
@@ -764,2 +764,32 @@ * Returns the sum of each row of the input tensor in the given dimension dim.
 
+    min(dim = null, keepdim = false) {
+        if (dim !== null) {
+            throw new Error("`dim !== null` not yet implemented.");
+        }
+        const value = min(this.data)[0];
+        return new Tensor(this.type, [value], []);
+    }
+    max(dim = null, keepdim = false) {
+        if (dim !== null) {
+            throw new Error("`dim !== null` not yet implemented.");
+        }
+        const value = max(this.data)[0];
+        return new Tensor(this.type, [value], []);
+    }
+
+    argmin(dim = null, keepdim = false) {
+        if (dim !== null) {
+            throw new Error("`dim !== null` not yet implemented.");
+        }
+        const index = min(this.data)[1];
+        return new Tensor('int64', [BigInt(index)], []);
+    }
+    argmax(dim = null, keepdim = false) {
+        if (dim !== null) {
+            throw new Error("`dim !== null` not yet implemented.");
+        }
+        const index = max(this.data)[1];
+        return new Tensor('int64', [BigInt(index)], []);
+    }
+
     /**
@@ -898,3 +928,3 @@ * Performs Tensor dtype conversion.
  * @param {[number, number]|[number, number, number]|[number, number, number, number]} [options.size=null] output spatial size.
- * @param {"bilinear"|"bicubic"} [options.mode='bilinear'] algorithm used for upsampling
+ * @param {"nearest"|"bilinear"|"bicubic"} [options.mode='bilinear'] algorithm used for upsampling
  * @returns {Promise<Tensor>} The interpolated tensor.
@@ -929,3 +959,5 @@ */
     let op;
-    if (mode === 'bilinear') {
+    if (mode === 'nearest') {
+        op = await TensorOpRegistry.nearest_interpolate_4d;
+    } else if (mode === 'bilinear') {
         op = await TensorOpRegistry.bilinear_interpolate_4d;
@@ -971,3 +1003,3 @@ } else if (mode === 'bicubic') {
  * @param {Tensor} x the input tensor
- * @param {number} k the k in "top-k"
+ * @param {number} [k] the k in "top-k"
  * @returns {Promise<[Tensor, Tensor]>} the output tuple of (Tensor, LongTensor) of top-k elements and their indices.
@@ -978,3 +1010,3 @@ */
 
-    if (k === null) {
+    if (k == null) {
         k = x.dims.at(-1);
@@ -1008,6 +1040,6 @@ } else {
     return await op({
-        x: data, 
-        s: arrayToIndexTensor(starts), 
-        e: arrayToIndexTensor(ends), 
-        a: arrayToIndexTensor(axes), 
+        x: data,
+        s: arrayToIndexTensor(starts),
+        e: arrayToIndexTensor(ends),
+        a: arrayToIndexTensor(axes),
         t: arrayToIndexTensor(steps ?? new Array(axes.length).fill(1)),
@@ -1014,0 +1046,0 @@ });

types/base/feature_extraction_utils.d.ts

@@ -17,15 +17,15 @@ /**
     /**
-     * Instantiate one of the processor classes of the library from a pretrained model.
+     * Instantiate one of the feature extractor 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)
+     * The feature extractor class to instantiate is selected based on the `feature_extractor_type` 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.
+     * - A string, the *model id* of a pretrained feature_extractor 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.
+     * - A path to a *directory* containing feature_extractor files, e.g., `./my_model_directory/`.
+     * @param {import('../utils/hub.js').PretrainedOptions} options Additional options for loading the feature_extractor.
      *
-     * @returns {Promise<FeatureExtractor>} A new instance of the Processor class.
+     * @returns {Promise<FeatureExtractor>} A new instance of the Feature Extractor class.
      */
@@ -32,0 +32,0 @@ static from_pretrained(pretrained_model_name_or_path: string, options: import("../utils/hub.js").PretrainedOptions): Promise<FeatureExtractor>;

types/base/processing_utils.d.ts

@@ -8,2 +8,3 @@ declare const Processor_base: new () => {
  * @typedef {import('../utils/hub.js').PretrainedOptions & ProcessorProperties} PretrainedProcessorOptions
+ * @typedef {import('../tokenizers.js').PreTrainedTokenizer} PreTrainedTokenizer
  */
@@ -19,4 +20,4 @@ /**
      *
-     * The processor class to instantiate is selected based on the `feature_extractor_type` property of the config object
-     * (either passed as an argument or loaded from `pretrained_model_name_or_path` if possible)
+     * 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)
      *
@@ -46,5 +47,5 @@ * @param {string} pretrained_model_name_or_path The name or path of the pretrained model. Can be either:
     /**
-     * @returns {import('../tokenizers.js').PreTrainedTokenizer|undefined} The tokenizer of the processor, if it exists.
+     * @returns {PreTrainedTokenizer|undefined} The tokenizer of the processor, if it exists.
      */
-    get tokenizer(): import("../tokenizers.js").PreTrainedTokenizer | undefined;
+    get tokenizer(): PreTrainedTokenizer | undefined;
     /**
@@ -54,18 +55,14 @@ * @returns {import('./feature_extraction_utils.js').FeatureExtractor|undefined} The feature extractor of the processor, if it exists.
     get feature_extractor(): import("./feature_extraction_utils.js").FeatureExtractor | undefined;
-    apply_chat_template(messages: any, options?: {}): string | number[] | number[][] | import("../transformers.js").Tensor | {
-        /**
-         * List of token ids to be fed to a model.
-         */
-        input_ids: number[] | number[][] | import("../transformers.js").Tensor;
-        /**
-         * List of indices specifying which tokens should be attended to by the model.
-         */
-        attention_mask: number[] | number[][] | import("../transformers.js").Tensor;
-        /**
-         * List of token type ids to be fed to a model.
-         */
-        token_type_ids?: number[] | number[][] | import("../transformers.js").Tensor;
-    };
-    batch_decode(...args: any[]): string[];
     /**
+     * @param {Parameters<PreTrainedTokenizer['apply_chat_template']>[0]} messages
+     * @param {Parameters<PreTrainedTokenizer['apply_chat_template']>[1]} options
+     * @returns {ReturnType<PreTrainedTokenizer['apply_chat_template']>}
+     */
+    apply_chat_template(messages: Parameters<PreTrainedTokenizer["apply_chat_template"]>[0], options?: Parameters<PreTrainedTokenizer["apply_chat_template"]>[1]): ReturnType<PreTrainedTokenizer["apply_chat_template"]>;
+    /**
+     * @param {Parameters<PreTrainedTokenizer['batch_decode']>} args
+     * @returns {ReturnType<PreTrainedTokenizer['batch_decode']>}
+     */
+    batch_decode(batch: number[][] | import("../transformers.js").Tensor, decode_args?: any): ReturnType<PreTrainedTokenizer["batch_decode"]>;
+    /**
      * Calls the feature_extractor function with the given input.
@@ -83,3 +80,4 @@ * @param {any} input The input to extract features from.
 export type PretrainedProcessorOptions = import("../utils/hub.js").PretrainedOptions & ProcessorProperties;
+export type PreTrainedTokenizer = import("../tokenizers.js").PreTrainedTokenizer;
 export {};
 //# sourceMappingURL=processing_utils.d.ts.map

types/generation/parameters.d.ts

@@ -1,2 +0,2 @@
-type GenerationFunctionParameters = {
+export type GenerationFunctionParameters = {
     /**
@@ -3,0 +3,0 @@ * (`Tensor` of varying shape depending on the modality, *optional*):

types/models/phi3_v/processing_phi3_v.d.ts

@@ -8,6 +8,10 @@ export class Phi3VProcessor extends Processor {
      * @param {RawImage|RawImage[]} images
-     * @param  {...any} args
+     * @param  { { padding?: boolean, truncation?: boolean, num_crops?: number } | undefined } options
      * @returns {Promise<any>}
      */
-    _call(text: string | string[], images?: RawImage | RawImage[], { padding, truncation, num_crops, }?: any[]): Promise<any>;
+    _call(text: string | string[], images?: RawImage | RawImage[], { padding, truncation, num_crops, }?: {
+        padding?: boolean;
+        truncation?: boolean;
+        num_crops?: number;
+    } | undefined): Promise<any>;
 }
@@ -14,0 +18,0 @@ import { Processor } from "../../base/processing_utils.js";

types/models/whisper/generation_whisper.d.ts

@@ -72,3 +72,3 @@ export class WhisperGenerationConfig extends GenerationConfig {
 }
-export type WhisperGenerationFunctionParameters = any & {
+export type WhisperGenerationFunctionParameters = import("../../generation/parameters.js").GenerationFunctionParameters & {
     generation_config: WhisperGenerationConfig;
@@ -75,0 +75,0 @@ } & WhisperGenerationConfig;

types/ops/registry.d.ts

 export class TensorOpRegistry {
     static session_options: {};
+    static get nearest_interpolate_4d(): Promise<(arg0: Record<string, Tensor>) => Promise<Tensor>>;
     static get bilinear_interpolate_4d(): Promise<(arg0: Record<string, Tensor>) => Promise<Tensor>>;
@@ -4,0 +5,0 @@ static get bicubic_interpolate_4d(): Promise<(arg0: Record<string, Tensor>) => Promise<Tensor>>;

types/utils/hub.d.ts

@@ -106,3 +106,3 @@ /**
     headers: Headers;
-    exists: any;
+    exists: boolean;
     status: number;
@@ -109,0 +109,0 @@ statusText: string;

types/utils/image.d.ts

@@ -132,4 +132,4 @@ export class RawImage {
      */
-    save(path: string): Promise<any>;
-    toSharp(): any;
+    save(path: string): Promise<sharp.OutputInfo>;
+    toSharp(): sharp.Sharp;
 }
@@ -141,2 +141,3 @@ /**
 import { Tensor } from './tensor.js';
+import sharp from 'sharp';
 //# sourceMappingURL=image.d.ts.map

types/utils/maths.d.ts

@@ -64,14 +64,16 @@ /**
  * Returns the value and index of the minimum element in an array.
- * @param {number[]|TypedArray} arr array of numbers.
- * @returns {[number, number]} the value and index of the minimum element, of the form: [valueOfMin, indexOfMin]
+ * @template {number[]|bigint[]|AnyTypedArray} T
+ * @param {T} arr array of numbers.
+ * @returns {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} the value and index of the minimum element, of the form: [valueOfMin, indexOfMin]
  * @throws {Error} If array is empty.
  */
-export function min(arr: number[] | TypedArray): [number, number];
+export function min<T extends number[] | bigint[] | AnyTypedArray>(arr: T): T extends bigint[] | BigTypedArray ? [bigint, number] : [number, number];
 /**
  * Returns the value and index of the maximum element in an array.
- * @param {number[]|AnyTypedArray} arr array of numbers.
- * @returns {[number, number]} the value and index of the maximum element, of the form: [valueOfMax, indexOfMax]
+ * @template {number[]|bigint[]|AnyTypedArray} T
+ * @param {T} arr array of numbers.
+ * @returns {T extends bigint[]|BigTypedArray ? [bigint, number] : [number, number]} the value and index of the maximum element, of the form: [valueOfMax, indexOfMax]
  * @throws {Error} If array is empty.
  */
-export function max(arr: number[] | AnyTypedArray): [number, number];
+export function max<T extends number[] | bigint[] | AnyTypedArray>(arr: T): T extends bigint[] | BigTypedArray ? [bigint, number] : [number, number];
 /**
@@ -78,0 +80,0 @@ * Performs median filter on the provided data. Padding is done by mirroring the data.

types/utils/tensor.d.ts

@@ -23,3 +23,3 @@ /**
  * @param {[number, number]|[number, number, number]|[number, number, number, number]} [options.size=null] output spatial size.
- * @param {"bilinear"|"bicubic"} [options.mode='bilinear'] algorithm used for upsampling
+ * @param {"nearest"|"bilinear"|"bicubic"} [options.mode='bilinear'] algorithm used for upsampling
  * @returns {Promise<Tensor>} The interpolated tensor.
@@ -29,3 +29,3 @@ */
     size?: [number, number] | [number, number, number] | [number, number, number, number];
-    mode?: "bilinear" | "bicubic";
+    mode?: "nearest" | "bilinear" | "bicubic";
 }): Promise<Tensor>;
@@ -52,6 +52,6 @@ /**
  * @param {Tensor} x the input tensor
- * @param {number} k the k in "top-k"
+ * @param {number} [k] the k in "top-k"
  * @returns {Promise<[Tensor, Tensor]>} the output tuple of (Tensor, LongTensor) of top-k elements and their indices.
  */
-export function topk(x: Tensor, k: number): Promise<[Tensor, Tensor]>;
+export function topk(x: Tensor, k?: number): Promise<[Tensor, Tensor]>;
 /**
@@ -434,2 +434,6 @@ * Slice a multidimensional float32 tensor.
     mean(dim?: any, keepdim?: boolean): Tensor;
+    min(dim?: any, keepdim?: boolean): Tensor;
+    max(dim?: any, keepdim?: boolean): Tensor;
+    argmin(dim?: any, keepdim?: boolean): Tensor;
+    argmax(dim?: any, keepdim?: boolean): Tensor;
     /**
@@ -436,0 +440,0 @@ * Performs Tensor dtype conversion.

No alert changes

Improved metrics

Total package byte prevSize

41426689

increased by0.16%

Number of package files

311

increased by0.97%

Lines of code

141623

increased by0.35%

No dependency changes