@huggingface/transformers 3.0.0-alpha.14 → 3.0.0-alpha.15

package/dist/transformers.cjs

@@ -3779,7 +3779,7 @@ __webpack_require__.r(__webpack_exports__);
 // Licensed under the MIT License.
 // This file is generated by /js/scripts/update-version.ts
 // Do not modify file content manually.
-const version = '1.19.0';
+const version = '1.19.2';
 //# sourceMappingURL=version.js.map
 
 /***/ }),
@@ -4437,7 +4437,7 @@ __webpack_require__.r(__webpack_exports__);
 
 
 
-const VERSION = '3.0.0-alpha.14';
+const VERSION = '3.0.0-alpha.15';
 
 // Check if various APIs are available (depends on environment)
 const IS_BROWSER_ENV = typeof self !== 'undefined';
@@ -4484,19 +4484,19 @@ const apis = Object.freeze({
 });
 
 const RUNNING_LOCALLY = IS_FS_AVAILABLE && IS_PATH_AVAILABLE;
-const __dirname = RUNNING_LOCALLY
+const dirname__ = RUNNING_LOCALLY
     ? path__WEBPACK_IMPORTED_MODULE_1__.dirname(path__WEBPACK_IMPORTED_MODULE_1__.dirname(url__WEBPACK_IMPORTED_MODULE_2__.fileURLToPath("file:///workspaces/transformers.js/src/env.js")))
     : './';
 
 // Only used for environments with access to file system
 const DEFAULT_CACHE_DIR = RUNNING_LOCALLY
-    ? path__WEBPACK_IMPORTED_MODULE_1__.join(__dirname, '/.cache/')
+    ? path__WEBPACK_IMPORTED_MODULE_1__.join(dirname__, '/.cache/')
     : null;
 
 // Set local model path, based on available APIs
 const DEFAULT_LOCAL_MODEL_PATH = '/models/';
 const localModelPath = RUNNING_LOCALLY
-    ? path__WEBPACK_IMPORTED_MODULE_1__.join(__dirname, DEFAULT_LOCAL_MODEL_PATH)
+    ? path__WEBPACK_IMPORTED_MODULE_1__.join(dirname__, DEFAULT_LOCAL_MODEL_PATH)
     : DEFAULT_LOCAL_MODEL_PATH;
 
 /**
@@ -6382,6 +6382,7 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   AutoModelForTextToSpectrogram: () => (/* binding */ AutoModelForTextToSpectrogram),
 /* harmony export */   AutoModelForTextToWaveform: () => (/* binding */ AutoModelForTextToWaveform),
 /* harmony export */   AutoModelForTokenClassification: () => (/* binding */ AutoModelForTokenClassification),
+/* harmony export */   AutoModelForUniversalSegmentation: () => (/* binding */ AutoModelForUniversalSegmentation),
 /* harmony export */   AutoModelForVision2Seq: () => (/* binding */ AutoModelForVision2Seq),
 /* harmony export */   AutoModelForXVector: () => (/* binding */ AutoModelForXVector),
 /* harmony export */   AutoModelForZeroShotObjectDetection: () => (/* binding */ AutoModelForZeroShotObjectDetection),
@@ -6413,7 +6414,9 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   CLIPSegForImageSegmentation: () => (/* binding */ CLIPSegForImageSegmentation),
 /* harmony export */   CLIPSegModel: () => (/* binding */ CLIPSegModel),
 /* harmony export */   CLIPSegPreTrainedModel: () => (/* binding */ CLIPSegPreTrainedModel),
+/* harmony export */   CLIPTextModel: () => (/* binding */ CLIPTextModel),
 /* harmony export */   CLIPTextModelWithProjection: () => (/* binding */ CLIPTextModelWithProjection),
+/* harmony export */   CLIPVisionModel: () => (/* binding */ CLIPVisionModel),
 /* harmony export */   CLIPVisionModelWithProjection: () => (/* binding */ CLIPVisionModelWithProjection),
 /* harmony export */   CamembertForMaskedLM: () => (/* binding */ CamembertForMaskedLM),
 /* harmony export */   CamembertForQuestionAnswering: () => (/* binding */ CamembertForQuestionAnswering),
@@ -6462,6 +6465,8 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   DebertaV2ForTokenClassification: () => (/* binding */ DebertaV2ForTokenClassification),
 /* harmony export */   DebertaV2Model: () => (/* binding */ DebertaV2Model),
 /* harmony export */   DebertaV2PreTrainedModel: () => (/* binding */ DebertaV2PreTrainedModel),
+/* harmony export */   DecisionTransformerModel: () => (/* binding */ DecisionTransformerModel),
+/* harmony export */   DecisionTransformerPreTrainedModel: () => (/* binding */ DecisionTransformerPreTrainedModel),
 /* harmony export */   DeiTForImageClassification: () => (/* binding */ DeiTForImageClassification),
 /* harmony export */   DeiTModel: () => (/* binding */ DeiTModel),
 /* harmony export */   DeiTPreTrainedModel: () => (/* binding */ DeiTPreTrainedModel),
@@ -6530,6 +6535,8 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   GemmaForCausalLM: () => (/* binding */ GemmaForCausalLM),
 /* harmony export */   GemmaModel: () => (/* binding */ GemmaModel),
 /* harmony export */   GemmaPreTrainedModel: () => (/* binding */ GemmaPreTrainedModel),
+/* harmony export */   GroupViTModel: () => (/* binding */ GroupViTModel),
+/* harmony export */   GroupViTPreTrainedModel: () => (/* binding */ GroupViTPreTrainedModel),
 /* harmony export */   HieraForImageClassification: () => (/* binding */ HieraForImageClassification),
 /* harmony export */   HieraModel: () => (/* binding */ HieraModel),
 /* harmony export */   HieraPreTrainedModel: () => (/* binding */ HieraPreTrainedModel),
@@ -6569,6 +6576,9 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   MarianMTModel: () => (/* binding */ MarianMTModel),
 /* harmony export */   MarianModel: () => (/* binding */ MarianModel),
 /* harmony export */   MarianPreTrainedModel: () => (/* binding */ MarianPreTrainedModel),
+/* harmony export */   MaskFormerForInstanceSegmentation: () => (/* binding */ MaskFormerForInstanceSegmentation),
+/* harmony export */   MaskFormerModel: () => (/* binding */ MaskFormerModel),
+/* harmony export */   MaskFormerPreTrainedModel: () => (/* binding */ MaskFormerPreTrainedModel),
 /* harmony export */   MaskedLMOutput: () => (/* binding */ MaskedLMOutput),
 /* harmony export */   MistralForCausalLM: () => (/* binding */ MistralForCausalLM),
 /* harmony export */   MistralModel: () => (/* binding */ MistralModel),
@@ -6627,6 +6637,9 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   PhiPreTrainedModel: () => (/* binding */ PhiPreTrainedModel),
 /* harmony export */   PreTrainedModel: () => (/* binding */ PreTrainedModel),
 /* harmony export */   PretrainedMixin: () => (/* binding */ PretrainedMixin),
+/* harmony export */   PvtForImageClassification: () => (/* binding */ PvtForImageClassification),
+/* harmony export */   PvtModel: () => (/* binding */ PvtModel),
+/* harmony export */   PvtPreTrainedModel: () => (/* binding */ PvtPreTrainedModel),
 /* harmony export */   PyAnnoteForAudioFrameClassification: () => (/* binding */ PyAnnoteForAudioFrameClassification),
 /* harmony export */   PyAnnoteModel: () => (/* binding */ PyAnnoteModel),
 /* harmony export */   PyAnnotePreTrainedModel: () => (/* binding */ PyAnnotePreTrainedModel),
@@ -6712,6 +6725,11 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   UniSpeechSatModel: () => (/* binding */ UniSpeechSatModel),
 /* harmony export */   UniSpeechSatPreTrainedModel: () => (/* binding */ UniSpeechSatPreTrainedModel),
 /* harmony export */   ViTForImageClassification: () => (/* binding */ ViTForImageClassification),
+/* harmony export */   ViTMAEModel: () => (/* binding */ ViTMAEModel),
+/* harmony export */   ViTMAEPreTrainedModel: () => (/* binding */ ViTMAEPreTrainedModel),
+/* harmony export */   ViTMSNForImageClassification: () => (/* binding */ ViTMSNForImageClassification),
+/* harmony export */   ViTMSNModel: () => (/* binding */ ViTMSNModel),
+/* harmony export */   ViTMSNPreTrainedModel: () => (/* binding */ ViTMSNPreTrainedModel),
 /* harmony export */   ViTModel: () => (/* binding */ ViTModel),
 /* harmony export */   ViTPreTrainedModel: () => (/* binding */ ViTPreTrainedModel),
 /* harmony export */   VisionEncoderDecoderModel: () => (/* binding */ VisionEncoderDecoderModel),
@@ -10227,6 +10245,18 @@ class CLIPPreTrainedModel extends PreTrainedModel { }
  */
 class CLIPModel extends CLIPPreTrainedModel { }
 
+/**
+ * The text model from CLIP without any head or projection on top.
+ */
+class CLIPTextModel extends CLIPPreTrainedModel {
+    /** @type {PreTrainedModel.from_pretrained} */
+    static async from_pretrained(pretrained_model_name_or_path, options = {}) {
+        // Update default model file name if not provided
+        options.model_file_name ??= 'text_model';
+        return super.from_pretrained(pretrained_model_name_or_path, options);
+    }
+}
+
 /**
  * CLIP Text Model with a projection layer on top (a linear layer on top of the pooled output)
  * 
@@ -10254,7 +10284,6 @@ class CLIPModel extends CLIPPreTrainedModel { }
  * ```
  */
 class CLIPTextModelWithProjection extends CLIPPreTrainedModel {
-
     /** @type {PreTrainedModel.from_pretrained} */
     static async from_pretrained(pretrained_model_name_or_path, options = {}) {
         // Update default model file name if not provided
@@ -10263,6 +10292,18 @@ class CLIPTextModelWithProjection extends CLIPPreTrainedModel {
     }
 }
 
+/**
+ * The vision model from CLIP without any head or projection on top.
+ */
+class CLIPVisionModel extends CLIPPreTrainedModel {
+    /** @type {PreTrainedModel.from_pretrained} */
+    static async from_pretrained(pretrained_model_name_or_path, options = {}) {
+        // Update default model file name if not provided
+        options.model_file_name ??= 'vision_model';
+        return super.from_pretrained(pretrained_model_name_or_path, options);
+    }
+}
+
 /**
  * CLIP Vision Model with a projection layer on top (a linear layer on top of the pooled output)
  * 
@@ -10929,6 +10970,43 @@ class ViTForImageClassification extends ViTPreTrainedModel {
 }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+class PvtPreTrainedModel extends PreTrainedModel { }
+class PvtModel extends PvtPreTrainedModel { }
+class PvtForImageClassification extends PvtPreTrainedModel {
+    /**
+     * @param {any} model_inputs
+     */
+    async _call(model_inputs) {
+        return new SequenceClassifierOutput(await super._call(model_inputs));
+    }
+}
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+class ViTMAEPreTrainedModel extends PreTrainedModel { }
+class ViTMAEModel extends ViTMAEPreTrainedModel { }
+//////////////////////////////////////////////////
+
+
+//////////////////////////////////////////////////
+class ViTMSNPreTrainedModel extends PreTrainedModel { }
+class ViTMSNModel extends ViTMSNPreTrainedModel { }
+class ViTMSNForImageClassification extends ViTMSNPreTrainedModel {
+    /**
+     * @param {any} model_inputs
+     */
+    async _call(model_inputs) {
+        return new SequenceClassifierOutput(await super._call(model_inputs));
+    }
+}
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+class GroupViTPreTrainedModel extends PreTrainedModel { }
+class GroupViTModel extends GroupViTPreTrainedModel { }
+//////////////////////////////////////////////////
+
 
 //////////////////////////////////////////////////
 class FastViTPreTrainedModel extends PreTrainedModel { }
@@ -11341,6 +11419,11 @@ class SapiensForDepthEstimation extends SapiensPreTrainedModel { }
 class SapiensForNormalEstimation extends SapiensPreTrainedModel { }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+class MaskFormerPreTrainedModel extends PreTrainedModel { }
+class MaskFormerModel extends MaskFormerPreTrainedModel { }
+class MaskFormerForInstanceSegmentation extends MaskFormerPreTrainedModel { }
+//////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
 class GLPNPreTrainedModel extends PreTrainedModel { }
@@ -12863,6 +12946,7 @@ class MusicgenForConditionalGeneration extends PreTrainedModel { // NOTE: not Mu
         return audio_values;
     }
 }
+//////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
 // MobileNetV1 models
@@ -12956,6 +13040,17 @@ class MobileNetV4ForImageClassification extends MobileNetV4PreTrainedModel {
 }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+// Decision Transformer models
+class DecisionTransformerPreTrainedModel extends PreTrainedModel { }
+
+/**
+ * The model builds upon the GPT2 architecture to perform autoregressive prediction of actions in an offline RL setting.
+ * Refer to the paper for more details: https://arxiv.org/abs/2106.01345
+ */
+class DecisionTransformerModel extends DecisionTransformerPreTrainedModel { }
+
+//////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
 // AutoModels, used to simplify construction of PreTrainedModels
@@ -12994,7 +13089,7 @@ class PretrainedMixin {
         session_options = {},
     } = {}) {
 
-        let options = {
+        const options = {
             progress_callback,
             config,
             cache_dir,
@@ -13013,7 +13108,7 @@ class PretrainedMixin {
             throw new Error("`MODEL_CLASS_MAPPINGS` not implemented for this type of `AutoClass`: " + this.name);
         }
 
-        for (let MODEL_CLASS_MAPPING of this.MODEL_CLASS_MAPPINGS) {
+        for (const MODEL_CLASS_MAPPING of this.MODEL_CLASS_MAPPINGS) {
             const modelInfo = MODEL_CLASS_MAPPING.get(options.config.model_type);
             if (!modelInfo) {
                 continue; // Item not found in this mapping
@@ -13068,6 +13163,10 @@ const MODEL_MAPPING_NAMES_ENCODER_ONLY = new Map([
     ['rt_detr', ['RTDetrModel', RTDetrModel]],
     ['table-transformer', ['TableTransformerModel', TableTransformerModel]],
     ['vit', ['ViTModel', ViTModel]],
+    ['pvt', ['PvtModel', PvtModel]],
+    ['vit_msn', ['ViTMSNModel', ViTMSNModel]],
+    ['vit_mae', ['ViTMAEModel', ViTMAEModel]],
+    ['groupvit', ['GroupViTModel', GroupViTModel]],
     ['fastvit', ['FastViTModel', FastViTModel]],
     ['mobilevit', ['MobileViTModel', MobileViTModel]],
     ['mobilevitv2', ['MobileViTV2Model', MobileViTV2Model]],
@@ -13090,10 +13189,14 @@ const MODEL_MAPPING_NAMES_ENCODER_ONLY = new Map([
     ['hifigan', ['SpeechT5HifiGan', SpeechT5HifiGan]],
     ['efficientnet', ['EfficientNetModel', EfficientNetModel]],
 
+    ['decision_transformer', ['DecisionTransformerModel', DecisionTransformerModel]],
+
     ['mobilenet_v1', ['MobileNetV1Model', MobileNetV1Model]],
     ['mobilenet_v2', ['MobileNetV2Model', MobileNetV2Model]],
     ['mobilenet_v3', ['MobileNetV3Model', MobileNetV3Model]],
     ['mobilenet_v4', ['MobileNetV4Model', MobileNetV4Model]],
+
+    ['maskformer', ['MaskFormerModel', MaskFormerModel]],
 ]);
 
 const MODEL_MAPPING_NAMES_ENCODER_DECODER = new Map([
@@ -13278,6 +13381,8 @@ const MODEL_FOR_DOCUMENT_QUESTION_ANSWERING_MAPPING_NAMES = new Map([
 
 const MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES = new Map([
     ['vit', ['ViTForImageClassification', ViTForImageClassification]],
+    ['pvt', ['PvtForImageClassification', PvtForImageClassification]],
+    ['vit_msn', ['ViTMSNForImageClassification', ViTMSNForImageClassification]],
     ['fastvit', ['FastViTForImageClassification', FastViTForImageClassification]],
     ['mobilevit', ['MobileViTForImageClassification', MobileViTForImageClassification]],
     ['mobilevitv2', ['MobileViTV2ForImageClassification', MobileViTV2ForImageClassification]],
@@ -13310,6 +13415,7 @@ const MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING_NAMES = new Map([
 ]);
 
 const MODEL_FOR_IMAGE_SEGMENTATION_MAPPING_NAMES = new Map([
+    // TODO: Do not add new models here
     ['detr', ['DetrForSegmentation', DetrForSegmentation]],
     ['clipseg', ['CLIPSegForImageSegmentation', CLIPSegForImageSegmentation]],
 ]);
@@ -13319,6 +13425,11 @@ const MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES = new Map([
     ['sapiens', ['SapiensForSemanticSegmentation', SapiensForSemanticSegmentation]],
 ]);
 
+const MODEL_FOR_UNIVERSAL_SEGMENTATION_MAPPING_NAMES = new Map([
+    ['detr', ['DetrForSegmentation', DetrForSegmentation]],
+    ['maskformer', ['MaskFormerForInstanceSegmentation', MaskFormerForInstanceSegmentation]],
+]);
+
 const MODEL_FOR_MASK_GENERATION_MAPPING_NAMES = new Map([
     ['sam', ['SamModel', SamModel]],
 ]);
@@ -13394,6 +13505,7 @@ const MODEL_CLASS_TYPE_MAPPING = [
     [MODEL_FOR_IMAGE_TEXT_TO_TEXT_MAPPING_NAMES, MODEL_TYPES.ImageTextToText],
     [MODEL_FOR_IMAGE_CLASSIFICATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_IMAGE_SEGMENTATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
+    [MODEL_FOR_UNIVERSAL_SEGMENTATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_IMAGE_MATTING_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_IMAGE_TO_IMAGE_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
@@ -13596,6 +13708,17 @@ class AutoModelForSemanticSegmentation extends PretrainedMixin {
     static MODEL_CLASS_MAPPINGS = [MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES];
 }
 
+/**
+ * Helper class which is used to instantiate pretrained universal image segmentation models with the `from_pretrained` function.
+ * The chosen model class is determined by the type specified in the model config.
+ * 
+ * @example
+ * let model = await AutoModelForUniversalSegmentation.from_pretrained('hf-internal-testing/tiny-random-MaskFormerForInstanceSegmentation');
+ */
+class AutoModelForUniversalSegmentation extends PretrainedMixin {
+    static MODEL_CLASS_MAPPINGS = [MODEL_FOR_UNIVERSAL_SEGMENTATION_MAPPING_NAMES];
+}
+
 /**
  * Helper class which is used to instantiate pretrained object detection models with the `from_pretrained` function.
  * The chosen model class is determined by the type specified in the model config.
@@ -17259,7 +17382,7 @@ const SUPPORTED_TASKS = Object.freeze({
     "image-segmentation": {
         // no tokenizer
         "pipeline": ImageSegmentationPipeline,
-        "model": [_models_js__WEBPACK_IMPORTED_MODULE_1__.AutoModelForImageSegmentation, _models_js__WEBPACK_IMPORTED_MODULE_1__.AutoModelForSemanticSegmentation],
+        "model": [_models_js__WEBPACK_IMPORTED_MODULE_1__.AutoModelForImageSegmentation, _models_js__WEBPACK_IMPORTED_MODULE_1__.AutoModelForSemanticSegmentation, _models_js__WEBPACK_IMPORTED_MODULE_1__.AutoModelForUniversalSegmentation],
         "processor": _processors_js__WEBPACK_IMPORTED_MODULE_2__.AutoProcessor,
         "default": {
             // TODO: replace with original
@@ -17501,7 +17624,7 @@ async function loadItems(mapping, model, pretrainedOptions) {
 
     /**@type {Promise[]} */
     const promises = [];
-    for (let [name, cls] of mapping.entries()) {
+    for (const [name, cls] of mapping.entries()) {
         if (!cls) continue;
 
         /**@type {Promise} */
@@ -17509,7 +17632,7 @@ async function loadItems(mapping, model, pretrainedOptions) {
         if (Array.isArray(cls)) {
             promise = new Promise(async (resolve, reject) => {
                 let e;
-                for (let c of cls) {
+                for (const c of cls) {
                     if (c === null) {
                         // If null, we resolve it immediately, meaning the relevant
                         // class was not found, but it is optional.
@@ -17547,7 +17670,7 @@ async function loadItems(mapping, model, pretrainedOptions) {
     await Promise.all(promises);
 
     // Then assign to result
-    for (let [name, promise] of Object.entries(result)) {
+    for (const [name, promise] of Object.entries(result)) {
         result[name] = await promise;
     }
 
@@ -17585,6 +17708,7 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   Florence2Processor: () => (/* binding */ Florence2Processor),
 /* harmony export */   GLPNFeatureExtractor: () => (/* binding */ GLPNFeatureExtractor),
 /* harmony export */   ImageFeatureExtractor: () => (/* binding */ ImageFeatureExtractor),
+/* harmony export */   MaskFormerFeatureExtractor: () => (/* binding */ MaskFormerFeatureExtractor),
 /* harmony export */   MobileNetV1FeatureExtractor: () => (/* binding */ MobileNetV1FeatureExtractor),
 /* harmony export */   MobileNetV2FeatureExtractor: () => (/* binding */ MobileNetV2FeatureExtractor),
 /* harmony export */   MobileNetV3FeatureExtractor: () => (/* binding */ MobileNetV3FeatureExtractor),
@@ -17596,6 +17720,7 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   OwlViTProcessor: () => (/* binding */ OwlViTProcessor),
 /* harmony export */   Owlv2ImageProcessor: () => (/* binding */ Owlv2ImageProcessor),
 /* harmony export */   Processor: () => (/* binding */ Processor),
+/* harmony export */   PvtImageProcessor: () => (/* binding */ PvtImageProcessor),
 /* harmony export */   PyAnnoteFeatureExtractor: () => (/* binding */ PyAnnoteFeatureExtractor),
 /* harmony export */   PyAnnoteProcessor: () => (/* binding */ PyAnnoteProcessor),
 /* harmony export */   RTDetrImageProcessor: () => (/* binding */ RTDetrImageProcessor),
@@ -17684,7 +17809,7 @@ function center_to_corners_format([centerX, centerY, width, height]) {
  * @param {Tensor} outputs.logits The logits
  * @param {Tensor} outputs.pred_boxes The predicted boxes.
  * @param {number} [threshold=0.5] The threshold to use for the scores.
- * @param {number[][]} [target_sizes=null] The sizes of the original images.
+ * @param {[number, number][]} [target_sizes=null] The sizes of the original images.
  * @param {boolean} [is_zero_shot=false] Whether zero-shot object detection was performed.
  * @return {Object[]} An array of objects containing the post-processed outputs.
  * @private
@@ -17765,7 +17890,7 @@ function post_process_object_detection(outputs, threshold = 0.5, target_sizes =
 /**
  * Post-processes the outputs of the model (for semantic segmentation).
  * @param {*} outputs Raw outputs of the model.
- * @param {number[][]} [target_sizes=null] List of tuples corresponding to the requested final size
+ * @param {[number, number][]} [target_sizes=null] List of tuples corresponding to the requested final size
  * (height, width) of each prediction. If unset, predictions will not be resized.
  * @returns {{segmentation: Tensor; labels: number[]}[]} The semantic segmentation maps.
  */
@@ -17825,6 +17950,300 @@ function post_process_semantic_segmentation(outputs, target_sizes = null) {
     return toReturn;
 }
 
+
+/**
+ * Binarize the given masks using `object_mask_threshold`, it returns the associated values of `masks`, `scores` and `labels`.
+ * @param {Tensor} class_logits The class logits.
+ * @param {Tensor} mask_logits The mask logits.
+ * @param {number} object_mask_threshold A number between 0 and 1 used to binarize the masks.
+ * @param {number} num_labels The number of labels.
+ * @returns {[Tensor[], number[], number[]]} The binarized masks, the scores, and the labels.
+ * @private
+ */
+function remove_low_and_no_objects(class_logits, mask_logits, object_mask_threshold, num_labels) {
+
+    const mask_probs_item = [];
+    const pred_scores_item = [];
+    const pred_labels_item = [];
+
+    for (let j = 0; j < class_logits.dims[0]; ++j) {
+        const cls = class_logits[j];
+        const mask = mask_logits[j];
+
+        const pred_label = (0,_utils_maths_js__WEBPACK_IMPORTED_MODULE_3__.max)(cls.data)[1];
+        if (pred_label === num_labels) {
+            // Is the background, so we ignore it
+            continue;
+        }
+
+        const scores = (0,_utils_maths_js__WEBPACK_IMPORTED_MODULE_3__.softmax)(cls.data);
+        const pred_score = scores[pred_label];
+        if (pred_score > object_mask_threshold) {
+            mask_probs_item.push(mask);
+            pred_scores_item.push(pred_score);
+            pred_labels_item.push(pred_label);
+        }
+    }
+
+    return [mask_probs_item, pred_scores_item, pred_labels_item];
+}
+
+/**
+ * Checks whether the segment is valid or not.
+ * @param {Int32Array} mask_labels Labels for each pixel in the mask.
+ * @param {Tensor[]} mask_probs Probabilities for each pixel in the masks.
+ * @param {number} k The class id of the segment.
+ * @param {number} mask_threshold The mask threshold.
+ * @param {number} overlap_mask_area_threshold The overlap mask area threshold.
+ * @returns {[boolean, number[]]} Whether the segment is valid or not, and the indices of the valid labels.
+ * @private
+ */
+function check_segment_validity(
+    mask_labels,
+    mask_probs,
+    k,
+    mask_threshold = 0.5,
+    overlap_mask_area_threshold = 0.8
+) {
+    // mask_k is a 1D array of indices, indicating where the mask is equal to k
+    const mask_k = [];
+    let mask_k_area = 0;
+    let original_area = 0;
+
+    const mask_probs_k_data = mask_probs[k].data;
+
+    // Compute the area of all the stuff in query k
+    for (let i = 0; i < mask_labels.length; ++i) {
+        if (mask_labels[i] === k) {
+            mask_k.push(i);
+            ++mask_k_area;
+        }
+
+        if (mask_probs_k_data[i] >= mask_threshold) {
+            ++original_area;
+        }
+    }
+    let mask_exists = mask_k_area > 0 && original_area > 0;
+
+    // Eliminate disconnected tiny segments
+    if (mask_exists) {
+        // Perform additional check
+        let area_ratio = mask_k_area / original_area;
+        mask_exists = area_ratio > overlap_mask_area_threshold;
+    }
+
+    return [mask_exists, mask_k]
+}
+
+/**
+ * Computes the segments.
+ * @param {Tensor[]} mask_probs The mask probabilities.
+ * @param {number[]} pred_scores The predicted scores.
+ * @param {number[]} pred_labels The predicted labels.
+ * @param {number} mask_threshold The mask threshold.
+ * @param {number} overlap_mask_area_threshold The overlap mask area threshold.
+ * @param {Set<number>} label_ids_to_fuse The label ids to fuse.
+ * @param {number[]} target_size The target size of the image.
+ * @returns {[Tensor, Array<{id: number, label_id: number, score: number}>]} The computed segments.
+ * @private
+ */
+function compute_segments(
+    mask_probs,
+    pred_scores,
+    pred_labels,
+    mask_threshold,
+    overlap_mask_area_threshold,
+    label_ids_to_fuse = null,
+    target_size = null,
+) {
+    const [height, width] = target_size ?? mask_probs[0].dims;
+
+    const segmentation = new _utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.Tensor(
+        'int32',
+        new Int32Array(height * width),
+        [height, width]
+    );
+    const segments = [];
+
+    // 1. If target_size is not null, we need to resize the masks to the target size
+    if (target_size !== null) {
+        // resize the masks to the target size
+        for (let i = 0; i < mask_probs.length; ++i) {
+            mask_probs[i] = (0,_utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.interpolate)(mask_probs[i], target_size, 'bilinear', false);
+        }
+    }
+
+    // 2. Weigh each mask by its prediction score
+    // NOTE: `mask_probs` is updated in-place
+    // 
+    // Temporary storage for the best label/scores for each pixel ([height, width]):
+    const mask_labels = new Int32Array(mask_probs[0].data.length);
+    const bestScores = new Float32Array(mask_probs[0].data.length);
+
+    for (let i = 0; i < mask_probs.length; ++i) {
+        let score = pred_scores[i];
+
+        const mask_probs_i_data = mask_probs[i].data;
+
+        for (let j = 0; j < mask_probs_i_data.length; ++j) {
+            mask_probs_i_data[j] *= score
+            if (mask_probs_i_data[j] > bestScores[j]) {
+                mask_labels[j] = i;
+                bestScores[j] = mask_probs_i_data[j];
+            }
+        }
+    }
+
+    let current_segment_id = 0;
+
+    // let stuff_memory_list = {}
+    const segmentation_data = segmentation.data;
+    for (let k = 0; k < pred_labels.length; ++k) {
+        const pred_class = pred_labels[k];
+
+        // TODO add `should_fuse`
+        // let should_fuse = pred_class in label_ids_to_fuse
+
+        // Check if mask exists and large enough to be a segment
+        const [mask_exists, mask_k] = check_segment_validity(
+            mask_labels,
+            mask_probs,
+            k,
+            mask_threshold,
+            overlap_mask_area_threshold
+        )
+
+        if (!mask_exists) {
+            // Nothing to see here
+            continue;
+        }
+
+        // TODO
+        // if (pred_class in stuff_memory_list) {
+        //     current_segment_id = stuff_memory_list[pred_class]
+        // } else {
+        //     current_segment_id += 1;
+        // }
+        ++current_segment_id;
+
+
+        // Add current object segment to final segmentation map
+        for (const index of mask_k) {
+            segmentation_data[index] = current_segment_id;
+        }
+
+        segments.push({
+            id: current_segment_id,
+            label_id: pred_class,
+            // was_fused: should_fuse, TODO
+            score: pred_scores[k],
+        })
+
+        // TODO
+        // if(should_fuse){
+        //     stuff_memory_list[pred_class] = current_segment_id
+        // }
+    }
+
+    return [segmentation, segments];
+}
+
+
+/**
+ * Post-process the model output to generate the final panoptic segmentation.
+ * @param {*} outputs The model output to post process
+ * @param {number} [threshold=0.5] The probability score threshold to keep predicted instance masks.
+ * @param {number} [mask_threshold=0.5] Threshold to use when turning the predicted masks into binary values.
+ * @param {number} [overlap_mask_area_threshold=0.8] The overlap mask area threshold to merge or discard small disconnected parts within each binary instance mask.
+ * @param {Set<number>} [label_ids_to_fuse=null] The labels in this state will have all their instances be fused together.
+ * @param {[number, number][]} [target_sizes=null] The target sizes to resize the masks to.
+ * @returns {Array<{ segmentation: Tensor, segments_info: Array<{id: number, label_id: number, score: number}>}>}
+ */
+function post_process_panoptic_segmentation(
+    outputs,
+    threshold = 0.5,
+    mask_threshold = 0.5,
+    overlap_mask_area_threshold = 0.8,
+    label_ids_to_fuse = null,
+    target_sizes = null,
+) {
+    if (label_ids_to_fuse === null) {
+        console.warn("`label_ids_to_fuse` unset. No instance will be fused.")
+        label_ids_to_fuse = new Set();
+    }
+
+    const class_queries_logits = outputs.class_queries_logits ?? outputs.logits; // [batch_size, num_queries, num_classes+1]
+    const masks_queries_logits = outputs.masks_queries_logits ?? outputs.pred_masks; // [batch_size, num_queries, height, width]
+
+    const mask_probs = masks_queries_logits.sigmoid()  // [batch_size, num_queries, height, width]
+
+    let [batch_size, num_queries, num_labels] = class_queries_logits.dims;
+    num_labels -= 1; // Remove last class (background)
+
+    if (target_sizes !== null && target_sizes.length !== batch_size) {
+        throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits")
+    }
+
+    let toReturn = [];
+    for (let i = 0; i < batch_size; ++i) {
+        let target_size = target_sizes !== null ? target_sizes[i] : null;
+
+        let class_logits = class_queries_logits[i];
+        let mask_logits = mask_probs[i];
+
+        let [mask_probs_item, pred_scores_item, pred_labels_item] = remove_low_and_no_objects(class_logits, mask_logits, threshold, num_labels);
+
+        if (pred_labels_item.length === 0) {
+            // No mask found
+            let [height, width] = target_size ?? mask_logits.dims.slice(-2);
+
+            let segmentation = new _utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.Tensor(
+                'int32',
+                new Int32Array(height * width).fill(-1),
+                [height, width]
+            )
+            toReturn.push({
+                segmentation: segmentation,
+                segments_info: []
+            });
+            continue;
+        }
+
+
+        // Get segmentation map and segment information of batch item
+        let [segmentation, segments] = compute_segments(
+            mask_probs_item,
+            pred_scores_item,
+            pred_labels_item,
+            mask_threshold,
+            overlap_mask_area_threshold,
+            label_ids_to_fuse,
+            target_size,
+        )
+
+        toReturn.push({
+            segmentation: segmentation,
+            segments_info: segments
+        })
+    }
+
+    return toReturn;
+}
+
+
+/**
+ * Post-processes the outputs of the model (for instance segmentation).
+ * @param {*} outputs Raw outputs of the model.
+ * @param {number} [threshold=0.5] The probability score threshold to keep predicted instance masks.
+ * @param {[number, number][]} [target_sizes=null] List of tuples corresponding to the requested final size
+ * (height, width) of each prediction. If unset, predictions will not be resized.
+ * @returns {Array<{ segmentation: Tensor, segments_info: Array<{id: number, label_id: number, score: number}>}>}
+ */
+function post_process_instance_segmentation(outputs, threshold = 0.5, target_sizes = null) {
+    throw new Error('Not implemented yet');
+    return [];
+}
+
 /**
  * Named tuple to indicate the order we are using is (height x width), even though
  * the Graphics’ industry standard is (width x height).
@@ -18413,6 +18832,7 @@ class SegformerFeatureExtractor extends ImageFeatureExtractor {
         return post_process_semantic_segmentation(...args);
     }
 }
+class PvtImageProcessor extends ImageFeatureExtractor { }
 class DPTFeatureExtractor extends ImageFeatureExtractor { }
 class DPTImageProcessor extends DPTFeatureExtractor { } // NOTE: extends DPTFeatureExtractor
 class BitImageProcessor extends ImageFeatureExtractor { }
@@ -18552,302 +18972,32 @@ class DetrFeatureExtractor extends ImageFeatureExtractor {
         // TODO support different mask sizes (not just 64x64)
         // Currently, just fill pixel mask with 1s
         const maskSize = [result.pixel_values.dims[0], 64, 64];
-        const pixel_mask = new _utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.Tensor(
-            'int64',
-            new BigInt64Array(maskSize.reduce((a, b) => a * b)).fill(1n),
-            maskSize
-        );
+        const pixel_mask = (0,_utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.full)(maskSize, 1n);
 
         return { ...result, pixel_mask };
     }
 
-    /**
-     * Post-processes the outputs of the model (for object detection).
-     * @param {Object} outputs The outputs of the model that must be post-processed
-     * @param {Tensor} outputs.logits The logits
-     * @param {Tensor} outputs.pred_boxes The predicted boxes.
-     * @return {Object[]} An array of objects containing the post-processed outputs.
-     */
-
     /** @type {typeof post_process_object_detection} */
     post_process_object_detection(...args) {
         return post_process_object_detection(...args);
     }
 
-    /**
-     * Binarize the given masks using `object_mask_threshold`, it returns the associated values of `masks`, `scores` and `labels`.
-     * @param {Tensor} class_logits The class logits.
-     * @param {Tensor} mask_logits The mask logits.
-     * @param {number} object_mask_threshold A number between 0 and 1 used to binarize the masks.
-     * @param {number} num_labels The number of labels.
-     * @returns {[Tensor[], number[], number[]]} The binarized masks, the scores, and the labels.
-     */
-    remove_low_and_no_objects(class_logits, mask_logits, object_mask_threshold, num_labels) {
-
-        let mask_probs_item = [];
-        let pred_scores_item = [];
-        let pred_labels_item = [];
-
-        for (let j = 0; j < class_logits.dims[0]; ++j) {
-            let cls = class_logits[j];
-            let mask = mask_logits[j];
-
-            let pred_label = (0,_utils_maths_js__WEBPACK_IMPORTED_MODULE_3__.max)(cls.data)[1];
-            if (pred_label === num_labels) {
-                // Is the background, so we ignore it
-                continue;
-            }
-
-            let scores = (0,_utils_maths_js__WEBPACK_IMPORTED_MODULE_3__.softmax)(cls.data);
-            let pred_score = scores[pred_label];
-            if (pred_score > object_mask_threshold) {
-                mask_probs_item.push(mask);
-                pred_scores_item.push(pred_score);
-                pred_labels_item.push(pred_label);
-            }
-        }
-
-        return [mask_probs_item, pred_scores_item, pred_labels_item];
-
-    }
-
-    /**
-     * Checks whether the segment is valid or not.
-     * @param {Int32Array} mask_labels Labels for each pixel in the mask.
-     * @param {Tensor[]} mask_probs Probabilities for each pixel in the masks.
-     * @param {number} k The class id of the segment.
-     * @param {number} mask_threshold The mask threshold.
-     * @param {number} overlap_mask_area_threshold The overlap mask area threshold.
-     * @returns {[boolean, number[]]} Whether the segment is valid or not, and the indices of the valid labels.
-     */
-    check_segment_validity(
-        mask_labels,
-        mask_probs,
-        k,
-        mask_threshold = 0.5,
-        overlap_mask_area_threshold = 0.8
-    ) {
-        // mask_k is a 1D array of indices, indicating where the mask is equal to k
-        let mask_k = [];
-        let mask_k_area = 0;
-        let original_area = 0;
-
-        const mask_probs_k_data = mask_probs[k].data;
-
-        // Compute the area of all the stuff in query k
-        for (let i = 0; i < mask_labels.length; ++i) {
-            if (mask_labels[i] === k) {
-                mask_k.push(i);
-                ++mask_k_area;
-            }
-
-            if (mask_probs_k_data[i] >= mask_threshold) {
-                ++original_area;
-            }
-        }
-        let mask_exists = mask_k_area > 0 && original_area > 0;
-
-        // Eliminate disconnected tiny segments
-        if (mask_exists) {
-            // Perform additional check
-            let area_ratio = mask_k_area / original_area;
-            mask_exists = area_ratio > overlap_mask_area_threshold;
-        }
-
-        return [mask_exists, mask_k]
+    /** @type {typeof post_process_panoptic_segmentation} */
+    post_process_panoptic_segmentation(...args) {
+        return post_process_panoptic_segmentation(...args);
     }
 
-    /**
-     * Computes the segments.
-     * @param {Tensor[]} mask_probs The mask probabilities.
-     * @param {number[]} pred_scores The predicted scores.
-     * @param {number[]} pred_labels The predicted labels.
-     * @param {number} mask_threshold The mask threshold.
-     * @param {number} overlap_mask_area_threshold The overlap mask area threshold.
-     * @param {Set<number>} label_ids_to_fuse The label ids to fuse.
-     * @param {number[]} target_size The target size of the image.
-     * @returns {[Tensor, Array<{id: number, label_id: number, score: number}>]} The computed segments.
-     */
-    compute_segments(
-        mask_probs,
-        pred_scores,
-        pred_labels,
-        mask_threshold,
-        overlap_mask_area_threshold,
-        label_ids_to_fuse = null,
-        target_size = null,
-    ) {
-        let [height, width] = target_size ?? mask_probs[0].dims;
-
-        let segmentation = new _utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.Tensor(
-            'int32',
-            new Int32Array(height * width),
-            [height, width]
-        );
-        let segments = [];
-
-        // 1. If target_size is not null, we need to resize the masks to the target size
-        if (target_size !== null) {
-            // resize the masks to the target size
-            for (let i = 0; i < mask_probs.length; ++i) {
-                mask_probs[i] = (0,_utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.interpolate)(mask_probs[i], target_size, 'bilinear', false);
-            }
-        }
-
-        // 2. Weigh each mask by its prediction score
-        // NOTE: `mask_probs` is updated in-place
-        // 
-        // Temporary storage for the best label/scores for each pixel ([height, width]):
-        let mask_labels = new Int32Array(mask_probs[0].data.length);
-        let bestScores = new Float32Array(mask_probs[0].data.length);
-
-        for (let i = 0; i < mask_probs.length; ++i) {
-            let score = pred_scores[i];
-
-            const mask_probs_i_data = mask_probs[i].data;
-
-            for (let j = 0; j < mask_probs_i_data.length; ++j) {
-                mask_probs_i_data[j] *= score
-                if (mask_probs_i_data[j] > bestScores[j]) {
-                    mask_labels[j] = i;
-                    bestScores[j] = mask_probs_i_data[j];
-                }
-            }
-        }
-
-        let current_segment_id = 0;
-
-        // let stuff_memory_list = {}
-        const segmentation_data = segmentation.data;
-        for (let k = 0; k < pred_labels.length; ++k) {
-            let pred_class = pred_labels[k];
-
-            // TODO add `should_fuse`
-            // let should_fuse = pred_class in label_ids_to_fuse
-
-            // Check if mask exists and large enough to be a segment
-            let [mask_exists, mask_k] = this.check_segment_validity(
-                mask_labels,
-                mask_probs,
-                k,
-                mask_threshold,
-                overlap_mask_area_threshold
-            )
-
-            if (!mask_exists) {
-                // Nothing to see here
-                continue;
-            }
-
-            // TODO
-            // if (pred_class in stuff_memory_list) {
-            //     current_segment_id = stuff_memory_list[pred_class]
-            // } else {
-            //     current_segment_id += 1;
-            // }
-            ++current_segment_id;
-
-
-            // Add current object segment to final segmentation map
-            for (let index of mask_k) {
-                segmentation_data[index] = current_segment_id;
-            }
-
-            segments.push({
-                id: current_segment_id,
-                label_id: pred_class,
-                // was_fused: should_fuse, TODO
-                score: pred_scores[k],
-            })
-
-            // TODO
-            // if(should_fuse){
-            //     stuff_memory_list[pred_class] = current_segment_id
-            // }
-        }
-
-        return [segmentation, segments];
+    post_process_instance_segmentation() {
+        // TODO
+        throw Error("Not implemented yet");
     }
+}
 
-    /**
-     * Post-process the model output to generate the final panoptic segmentation.
-     * @param {*} outputs The model output to post process
-     * @param {number} [threshold=0.5] The probability score threshold to keep predicted instance masks.
-     * @param {number} [mask_threshold=0.5] Threshold to use when turning the predicted masks into binary values.
-     * @param {number} [overlap_mask_area_threshold=0.8] The overlap mask area threshold to merge or discard small disconnected parts within each binary instance mask.
-     * @param {Set<number>} [label_ids_to_fuse=null] The labels in this state will have all their instances be fused together.
-     * @param {number[][]} [target_sizes=null] The target sizes to resize the masks to.
-     * @returns {Array<{ segmentation: Tensor, segments_info: Array<{id: number, label_id: number, score: number}>}>}
-     */
-    post_process_panoptic_segmentation(
-        outputs,
-        threshold = 0.5,
-        mask_threshold = 0.5,
-        overlap_mask_area_threshold = 0.8,
-        label_ids_to_fuse = null,
-        target_sizes = null,
-    ) {
-        if (label_ids_to_fuse === null) {
-            console.warn("`label_ids_to_fuse` unset. No instance will be fused.")
-            label_ids_to_fuse = new Set();
-        }
-
-        const class_queries_logits = outputs.logits; // [batch_size, num_queries, num_classes+1]
-        const masks_queries_logits = outputs.pred_masks; // [batch_size, num_queries, height, width]
-
-        const mask_probs = masks_queries_logits.sigmoid()  // [batch_size, num_queries, height, width]
-
-        let [batch_size, num_queries, num_labels] = class_queries_logits.dims;
-        num_labels -= 1; // Remove last class (background)
-
-        if (target_sizes !== null && target_sizes.length !== batch_size) {
-            throw Error("Make sure that you pass in as many target sizes as the batch dimension of the logits")
-        }
-
-        let toReturn = [];
-        for (let i = 0; i < batch_size; ++i) {
-            let target_size = target_sizes !== null ? target_sizes[i] : null;
-
-            let class_logits = class_queries_logits[i];
-            let mask_logits = mask_probs[i];
-
-            let [mask_probs_item, pred_scores_item, pred_labels_item] = this.remove_low_and_no_objects(class_logits, mask_logits, threshold, num_labels);
-
-            if (pred_labels_item.length === 0) {
-                // No mask found
-                let [height, width] = target_size ?? mask_logits.dims.slice(-2);
-
-                let segmentation = new _utils_tensor_js__WEBPACK_IMPORTED_MODULE_4__.Tensor(
-                    'int32',
-                    new Int32Array(height * width).fill(-1),
-                    [height, width]
-                )
-                toReturn.push({
-                    segmentation: segmentation,
-                    segments_info: []
-                });
-                continue;
-            }
-
-
-            // Get segmentation map and segment information of batch item
-            let [segmentation, segments] = this.compute_segments(
-                mask_probs_item,
-                pred_scores_item,
-                pred_labels_item,
-                mask_threshold,
-                overlap_mask_area_threshold,
-                label_ids_to_fuse,
-                target_size,
-            )
-
-            toReturn.push({
-                segmentation: segmentation,
-                segments_info: segments
-            })
-        }
+class MaskFormerFeatureExtractor extends ImageFeatureExtractor {
 
-        return toReturn;
+    /** @type {typeof post_process_panoptic_segmentation} */
+    post_process_panoptic_segmentation(...args) {
+        return post_process_panoptic_segmentation(...args);
     }
 
     post_process_instance_segmentation() {
@@ -18856,6 +19006,7 @@ class DetrFeatureExtractor extends ImageFeatureExtractor {
     }
 }
 
+
 class YolosFeatureExtractor extends ImageFeatureExtractor {
     /** @type {typeof post_process_object_detection} */
     post_process_object_detection(...args) {
@@ -20145,11 +20296,13 @@ class AutoProcessor {
         BitImageProcessor,
         DPTImageProcessor,
         DPTFeatureExtractor,
+        PvtImageProcessor,
         GLPNFeatureExtractor,
         BeitFeatureExtractor,
         DeiTFeatureExtractor,
         DetrFeatureExtractor,
         RTDetrImageProcessor,
+        MaskFormerFeatureExtractor,
         YolosFeatureExtractor,
         DonutFeatureExtractor,
         NougatImageProcessor,
@@ -20527,23 +20680,26 @@ function is_chinese_char(cp) {
 }
 
 /**
- * Helper function to fuse consecutive values in an array equal to the specified value.
- * @param {string[]} arr The input array
- * @param {any} value The value to fuse on.
- * @param {Map<string, any>} mapping The mapping from input domain to value.
+ * Helper function to fuse consecutive unknown tokens.
+ * @param {string[]} arr The list of input tokens
+ * @param {Map<string, any>} tokens_to_ids The mapping from tokens to token ids.
+ * @param {number} unk_token_id The value to fuse on.
+ * @private
  */
-function fuse(arr, value, mapping) {
+function fuse_unk(arr, tokens_to_ids, unk_token_id) {
     const fused = [];
     let i = 0;
     while (i < arr.length) {
         fused.push(arr[i])
-        if ((mapping.get(arr[i]) ?? value) !== value) {
+        if ((tokens_to_ids.get(arr[i]) ?? unk_token_id) !== unk_token_id) {
             ++i;
             continue;
         }
 
-        while (i < arr.length && (mapping.get(arr[i]) ?? value) === value) {
-            ++i;
+        while (++i < arr.length && (tokens_to_ids.get(arr[i]) ?? unk_token_id) === unk_token_id) {
+            if (tokens_to_ids.get(fused.at(-1)) !== unk_token_id) {
+                fused[fused.length - 1] += arr[i];
+            }
         }
     }
 
@@ -20660,15 +20816,15 @@ class TokenizerModel extends _utils_generic_js__WEBPACK_IMPORTED_MODULE_0__.Call
     /**
      * Internal function to call the TokenizerModel instance.
      * @param {string[]} tokens The tokens to encode.
-     * @returns {string[]} The encoded token IDs.
+     * @returns {string[]} The encoded tokens.
      */
     _call(tokens) {
-        let ids = this.encode(tokens);
+        tokens = this.encode(tokens);
         if (this.fuse_unk) {
             // Fuse unknown tokens
-            ids = fuse(ids, this.unk_token_id, this.tokens_to_ids);
+            tokens = fuse_unk(tokens, this.tokens_to_ids, this.unk_token_id);
         }
-        return ids;
+        return tokens;
     }
 
     /**
@@ -21166,15 +21322,19 @@ class BPE extends TokenizerModel {
             for (const t of bpe_token_list) {
                 if (this.tokens_to_ids.has(t)) {
                     outputTokens.push(t);
-                } else {
-                    if (this.byte_fallback) {
-                        outputTokens.push(
-                            ...Array.from(this.text_encoder.encode(t))
-                                .map(x => `<0x${x.toString(16).toUpperCase().padStart(2, '0')}>`)
-                        );
+                } else if (this.byte_fallback) {
+                    const byteTokens = Array.from(this.text_encoder.encode(t))
+                        .map(x => `<0x${x.toString(16).toUpperCase().padStart(2, '0')}>`);
+                    if (byteTokens.every(x => this.tokens_to_ids.has(x))) {
+                        // Ensure the byte tokens are actually in the vocabulary, otherwise
+                        // we fall back to the unknown token. For more information, see
+                        // https://github.com/huggingface/transformers/issues/28096.
+                        outputTokens.push(...byteTokens);
                     } else {
                         outputTokens.push(this.unk_token);
                     }
+                } else {
+                    outputTokens.push(this.unk_token);
                 }
             }
         }
@@ -22846,11 +23006,10 @@ class PreTrainedTokenizer extends _utils_generic_js__WEBPACK_IMPORTED_MODULE_0__
             this.decoder.end_of_word_suffix = this.model.end_of_word_suffix;
         }
 
-
         this.added_tokens_regex = this.added_tokens.length > 0 ? new RegExp(
-            this.added_tokens
+            this.added_tokens.slice()
                 // Sort by length (desc) to avoid early partial matches
-                .toSorted((a, b) => b.content.length - a.content.length)
+                .sort((a, b) => b.content.length - a.content.length)
                 .map(x => `${x.lstrip ? '\\s*' : ''}(${(0,_utils_core_js__WEBPACK_IMPORTED_MODULE_1__.escapeRegExp)(x.content)})${x.rstrip ? '\\s*' : ''}`)
                 .join('|')
         ) : null;
@@ -23348,6 +23507,67 @@ class PreTrainedTokenizer extends _utils_generic_js__WEBPACK_IMPORTED_MODULE_0__
 
         return decoded;
     }
+
+    /**
+     * Retrieve the chat template string used for tokenizing chat messages. This template is used
+     * internally by the `apply_chat_template` method and can also be used externally to retrieve the model's chat
+     * template for better generation tracking.
+     * 
+     * @param {Object} options An optional object containing the following properties:
+     * @param {string} [options.chat_template=null]
+     * A Jinja template or the name of a template to use for this conversion.
+     * It is usually not necessary to pass anything to this argument,
+     * as the model's template will be used by default.
+     * @param {Object[]} [options.tools=null]
+     * A list of tools (callable functions) that will be accessible to the model. If the template does not
+     * support function calling, this argument will have no effect. Each tool should be passed as a JSON Schema,
+     * giving the name, description and argument types for the tool. See our
+     * [chat templating guide](https://huggingface.co/docs/transformers/main/en/chat_templating#automated-function-conversion-for-tool-use)
+     * for more information.
+     * @returns {string} The chat template string.
+     */
+    get_chat_template({
+        chat_template = null,
+        tools = null,
+    } = {}) {
+
+        // First, handle the cases when the model has a dict of multiple templates
+        if (this.chat_template && typeof this.chat_template === 'object') {
+            const template_dict = this.chat_template;
+
+            if (chat_template !== null && Object.hasOwn(template_dict, chat_template)) {
+                // The user can pass the name of a template to the chat template argument instead of an entire template
+                chat_template = template_dict[chat_template];
+            } else if (chat_template === null) {
+                if (tools !== null && 'tool_use' in template_dict) {
+                    chat_template = template_dict['tool_use'];
+                } else if ('default' in template_dict) {
+                    chat_template = template_dict['default'];
+                } else {
+                    throw Error(
+                        `This model has multiple chat templates with no default specified! Please either pass a chat ` +
+                        `template or the name of the template you wish to use to the 'chat_template' argument. Available ` +
+                        `template names are ${Object.keys(template_dict).sort()}.`
+                    )
+                }
+            }
+        } else if (chat_template === null) {
+            // These are the cases when the model has a single template
+            // priority: `chat_template` argument > `tokenizer.chat_template`
+            if (this.chat_template) {
+                chat_template = this.chat_template;
+            } else {
+                throw Error(
+                    "Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template " +
+                    "argument was passed! For information about writing templates and setting the " +
+                    "tokenizer.chat_template attribute, please see the documentation at " +
+                    "https://huggingface.co/docs/transformers/main/en/chat_templating"
+                )
+            }
+        }
+        return chat_template;
+    }
+
     /**
      * Converts a list of message objects with `"role"` and `"content"` keys to a list of token
      * ids. This method is intended for use with chat models, and will read the tokenizer's chat_template attribute to
@@ -23421,39 +23641,8 @@ class PreTrainedTokenizer extends _utils_generic_js__WEBPACK_IMPORTED_MODULE_0__
         ...kwargs
     } = {}) {
 
-        // First, handle the cases when the model has a dict of multiple templates
-        if (
-            (this.chat_template && typeof this.chat_template === 'object')
-            || this.chat_template === null
-        ) {
-            const template_dict = this.chat_template;
+        chat_template = this.get_chat_template({ chat_template, tools });
 
-            if (chat_template !== null && Object.hasOwn(template_dict, chat_template)) {
-                // The user can pass the name of a template to the chat template argument instead of an entire template
-                chat_template = template_dict[chat_template];
-            } else if (chat_template === null && 'default' in template_dict) {
-                chat_template = template_dict['default'];
-            } else if (chat_template === null) {
-                throw Error(
-                    `This model has multiple chat templates with no default specified! Please either pass a chat ` +
-                    `template or the name of the template you wish to use to the 'chat_template' argument. Available ` +
-                    `template names are ${Object.keys(template_dict).sort()}.`
-                )
-            }
-        } else {
-            // These are the cases when the model has a single template
-            // priority: `chat_template` argument > `tokenizer.chat_template`
-            if (this.chat_template) {
-                chat_template = this.chat_template;
-            } else {
-                throw Error(
-                    "Cannot use apply_chat_template() because tokenizer.chat_template is not set and no template " +
-                    "argument was passed! For information about writing templates and setting the " +
-                    "tokenizer.chat_template attribute, please see the documentation at " +
-                    "https://huggingface.co/docs/transformers/main/en/chat_templating"
-                )
-            }
-        }
         if (typeof chat_template !== 'string') {
             throw Error(`chat_template must be a string, but got ${typeof chat_template}`);
         }
@@ -27871,11 +28060,20 @@ function softmax(arr) {
  * @returns {T} The resulting log_softmax array.
  */
 function log_softmax(arr) {
-    // Compute the softmax values
-    const softmaxArr = softmax(arr);
+    // Compute the maximum value in the array
+    const maxVal = max(arr)[0];
+
+    // Compute the sum of the exponentials
+    let sumExps = 0;
+    for(let i = 0; i < arr.length; ++i) {
+        sumExps += Math.exp(arr[i] - maxVal);
+    }
+
+    // Compute the log of the sum
+    const logSum = Math.log(sumExps);
 
-    // Apply log formula to each element
-    const logSoftmaxArr = softmaxArr.map(x => Math.log(x));
+    // Compute the softmax values
+    const logSoftmaxArr = arr.map(x => x - maxVal - logSum);
 
     return /** @type {T} */(logSoftmaxArr);
 }
@@ -30329,6 +30527,7 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   AutoModelForTextToSpectrogram: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForTextToSpectrogram),
 /* harmony export */   AutoModelForTextToWaveform: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForTextToWaveform),
 /* harmony export */   AutoModelForTokenClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForTokenClassification),
+/* harmony export */   AutoModelForUniversalSegmentation: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForUniversalSegmentation),
 /* harmony export */   AutoModelForVision2Seq: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForVision2Seq),
 /* harmony export */   AutoModelForXVector: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForXVector),
 /* harmony export */   AutoModelForZeroShotObjectDetection: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.AutoModelForZeroShotObjectDetection),
@@ -30373,8 +30572,10 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   CLIPSegForImageSegmentation: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPSegForImageSegmentation),
 /* harmony export */   CLIPSegModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPSegModel),
 /* harmony export */   CLIPSegPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPSegPreTrainedModel),
+/* harmony export */   CLIPTextModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPTextModel),
 /* harmony export */   CLIPTextModelWithProjection: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPTextModelWithProjection),
 /* harmony export */   CLIPTokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.CLIPTokenizer),
+/* harmony export */   CLIPVisionModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPVisionModel),
 /* harmony export */   CLIPVisionModelWithProjection: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CLIPVisionModelWithProjection),
 /* harmony export */   CamembertForMaskedLM: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CamembertForMaskedLM),
 /* harmony export */   CamembertForQuestionAnswering: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.CamembertForQuestionAnswering),
@@ -30436,6 +30637,8 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   DebertaV2Model: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DebertaV2Model),
 /* harmony export */   DebertaV2PreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DebertaV2PreTrainedModel),
 /* harmony export */   DebertaV2Tokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.DebertaV2Tokenizer),
+/* harmony export */   DecisionTransformerModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DecisionTransformerModel),
+/* harmony export */   DecisionTransformerPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DecisionTransformerPreTrainedModel),
 /* harmony export */   DeiTFeatureExtractor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.DeiTFeatureExtractor),
 /* harmony export */   DeiTForImageClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DeiTForImageClassification),
 /* harmony export */   DeiTModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.DeiTModel),
@@ -30525,6 +30728,8 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   GemmaPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.GemmaPreTrainedModel),
 /* harmony export */   GemmaTokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.GemmaTokenizer),
 /* harmony export */   Grok1Tokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.Grok1Tokenizer),
+/* harmony export */   GroupViTModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.GroupViTModel),
+/* harmony export */   GroupViTPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.GroupViTPreTrainedModel),
 /* harmony export */   HerbertTokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.HerbertTokenizer),
 /* harmony export */   HieraForImageClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.HieraForImageClassification),
 /* harmony export */   HieraModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.HieraModel),
@@ -30578,6 +30783,10 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   MarianModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MarianModel),
 /* harmony export */   MarianPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MarianPreTrainedModel),
 /* harmony export */   MarianTokenizer: () => (/* reexport safe */ _tokenizers_js__WEBPACK_IMPORTED_MODULE_3__.MarianTokenizer),
+/* harmony export */   MaskFormerFeatureExtractor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.MaskFormerFeatureExtractor),
+/* harmony export */   MaskFormerForInstanceSegmentation: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MaskFormerForInstanceSegmentation),
+/* harmony export */   MaskFormerModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MaskFormerModel),
+/* harmony export */   MaskFormerPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MaskFormerPreTrainedModel),
 /* harmony export */   MaskedLMOutput: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MaskedLMOutput),
 /* harmony export */   MaxLengthCriteria: () => (/* reexport safe */ _generation_stopping_criteria_js__WEBPACK_IMPORTED_MODULE_11__.MaxLengthCriteria),
 /* harmony export */   MistralForCausalLM: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.MistralForCausalLM),
@@ -30655,6 +30864,10 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   PretrainedConfig: () => (/* reexport safe */ _configs_js__WEBPACK_IMPORTED_MODULE_5__.PretrainedConfig),
 /* harmony export */   PretrainedMixin: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PretrainedMixin),
 /* harmony export */   Processor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.Processor),
+/* harmony export */   PvtForImageClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PvtForImageClassification),
+/* harmony export */   PvtImageProcessor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.PvtImageProcessor),
+/* harmony export */   PvtModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PvtModel),
+/* harmony export */   PvtPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PvtPreTrainedModel),
 /* harmony export */   PyAnnoteFeatureExtractor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.PyAnnoteFeatureExtractor),
 /* harmony export */   PyAnnoteForAudioFrameClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PyAnnoteForAudioFrameClassification),
 /* harmony export */   PyAnnoteModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.PyAnnoteModel),
@@ -30775,6 +30988,11 @@ __webpack_require__.r(__webpack_exports__);
 /* harmony export */   ViTFeatureExtractor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.ViTFeatureExtractor),
 /* harmony export */   ViTForImageClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTForImageClassification),
 /* harmony export */   ViTImageProcessor: () => (/* reexport safe */ _processors_js__WEBPACK_IMPORTED_MODULE_4__.ViTImageProcessor),
+/* harmony export */   ViTMAEModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTMAEModel),
+/* harmony export */   ViTMAEPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTMAEPreTrainedModel),
+/* harmony export */   ViTMSNForImageClassification: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTMSNForImageClassification),
+/* harmony export */   ViTMSNModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTMSNModel),
+/* harmony export */   ViTMSNPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTMSNPreTrainedModel),
 /* harmony export */   ViTModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTModel),
 /* harmony export */   ViTPreTrainedModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.ViTPreTrainedModel),
 /* harmony export */   VisionEncoderDecoderModel: () => (/* reexport safe */ _models_js__WEBPACK_IMPORTED_MODULE_2__.VisionEncoderDecoderModel),