@huggingface/transformers 3.0.0-alpha.9 → 3.0.1

package/src/models.js

@@ -71,6 +71,10 @@ import {
     getModelJSON,
 } from './utils/hub.js';
 
+import {
+    GITHUB_ISSUE_URL,
+} from './utils/constants.js';
+
 import {
     LogitsProcessorList,
     ForcedBOSTokenLogitsProcessor,
@@ -142,11 +146,12 @@ const MODEL_CLASS_TO_NAME_MAPPING = new Map();
  * @param {string} pretrained_model_name_or_path The path to the directory containing the model file.
  * @param {string} fileName The name of the model file.
  * @param {import('./utils/hub.js').PretrainedModelOptions} options Additional options for loading the model.
- * @returns {Promise<{buffer: Uint8Array, session_options: Object}>} A Promise that resolves to the data needed to create an InferenceSession object.
+ * @returns {Promise<{buffer: Uint8Array, session_options: Object, session_config: Object}>} A Promise that resolves to the data needed to create an InferenceSession object.
  * @private
  */
 async function getSession(pretrained_model_name_or_path, fileName, options) {
-    let device = options.device;
+    const custom_config = options.config?.['transformers.js_config'] ?? {};
+    let device = options.device ?? custom_config.device;
     if (device && typeof device !== 'string') {
         if (device.hasOwnProperty(fileName)) {
             device = device[fileName];
@@ -164,7 +169,7 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
 
     // If options.dtype is specified, we use it to choose the suffix for the model file.
     // Otherwise, we use the default dtype for the device.
-    let dtype = options.dtype;
+    let dtype = options.dtype ?? custom_config.dtype;
     if (typeof dtype !== 'string') {
         if (dtype && dtype.hasOwnProperty(fileName)) {
             dtype = dtype[fileName];
@@ -182,27 +187,54 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
         throw new Error(`The device (${selectedDevice}) does not support fp16.`);
     }
 
+    // Only valid for models with a decoder
+    const kv_cache_dtype = custom_config.kv_cache_dtype
+        ? (typeof custom_config.kv_cache_dtype === 'string'
+            ? custom_config.kv_cache_dtype
+            : custom_config.kv_cache_dtype[selectedDtype] ?? 'float32')
+        : undefined;
+
+    if (kv_cache_dtype && !['float32', 'float16'].includes(kv_cache_dtype)) {
+        throw new Error(`Invalid kv_cache_dtype: ${kv_cache_dtype}. Should be one of: float32, float16`);
+    }
+
+    const session_config = {
+        dtype: selectedDtype,
+        kv_cache_dtype,
+    }
+
     // Construct the model file name
     const suffix = DEFAULT_DTYPE_SUFFIX_MAPPING[selectedDtype];
     const modelFileName = `${options.subfolder ?? ''}/${fileName}${suffix}.onnx`;
 
-    const session_options = { ...options.session_options } ?? {};
+    const session_options = { ...options.session_options };
 
     // Overwrite `executionProviders` if not specified
     session_options.executionProviders ??= executionProviders;
 
+    // Overwrite `freeDimensionOverrides` if specified in config and not set in session options
+    const free_dimension_overrides = custom_config.free_dimension_overrides;
+    if (free_dimension_overrides) {
+        session_options.freeDimensionOverrides ??= free_dimension_overrides;
+    } else if (selectedDevice.startsWith('webnn') && !session_options.freeDimensionOverrides) {
+        console.warn(
+            'WebNN does not currently support dynamic shapes and requires `free_dimension_overrides` to be set in config.json as a field within "transformers.js_config". ' +
+            'When `free_dimension_overrides` is not set, you may experience significant performance degradation.'
+        );
+    }
 
     const bufferPromise = getModelFile(pretrained_model_name_or_path, modelFileName, true, options);
 
     // handle onnx external data files
+    const use_external_data_format = options.use_external_data_format ?? custom_config.use_external_data_format;
     /** @type {Promise<{path: string, data: Uint8Array}>[]} */
     let externalDataPromises = [];
-    if (options.use_external_data_format && (
-        options.use_external_data_format === true ||
+    if (use_external_data_format && (
+        use_external_data_format === true ||
         (
-            typeof options.use_external_data_format === 'object' &&
-            options.use_external_data_format.hasOwnProperty(fileName) &&
-            options.use_external_data_format[fileName] === true
+            typeof use_external_data_format === 'object' &&
+            use_external_data_format.hasOwnProperty(fileName) &&
+            use_external_data_format[fileName] === true
         )
     )) {
         if (apis.IS_NODE_ENV) {
@@ -236,6 +268,7 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
         });
         if (Object.keys(shapes).length > 0 && !isONNXProxy()) {
             // Only set preferredOutputLocation if shapes are present and we aren't proxying ONNX
+            /** @type {Record<string, import('onnxruntime-common').Tensor.DataLocation>} */
             const preferredOutputLocation = {};
             for (const key in shapes) {
                 preferredOutputLocation[key] = 'gpu-buffer';
@@ -245,7 +278,8 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
     }
 
     const buffer = await bufferPromise;
-    return { buffer, session_options };
+
+    return { buffer, session_options, session_config };
 }
 
 /**
@@ -260,13 +294,30 @@ async function getSession(pretrained_model_name_or_path, fileName, options) {
 async function constructSessions(pretrained_model_name_or_path, names, options) {
     return Object.fromEntries(await Promise.all(
         Object.keys(names).map(async (name) => {
-            const { buffer, session_options } = await getSession(pretrained_model_name_or_path, names[name], options);
-            const session = await createInferenceSession(buffer, session_options);
+            const { buffer, session_options, session_config } = await getSession(pretrained_model_name_or_path, names[name], options);
+            const session = await createInferenceSession(buffer, session_options, session_config);
             return [name, session];
         })
     ));
 }
 
+/**
+ * Helper function to load multiple optional configuration files
+ * @param {string} pretrained_model_name_or_path The path to the directory containing the config file.
+ * @param {Record<string, string>} names The names of the config files to load.
+ * @param {import('./utils/hub.js').PretrainedModelOptions} options Additional options for loading the configs.
+ * @returns {Promise<Record<string, any>>} A Promise that resolves to a dictionary of configuration objects.
+ * @private
+ */
+async function getOptionalConfigs(pretrained_model_name_or_path, names, options) {
+    return Object.fromEntries(await Promise.all(
+        Object.keys(names).map(async (name) => {
+            const config = await getModelJSON(pretrained_model_name_or_path, names[name], false, options);
+            return [name, config];
+        })
+    ));
+}
+
 /**
  * Validate model inputs
  * @param {Object} session The InferenceSession object that will be run.
@@ -360,7 +411,7 @@ function replaceTensors(obj) {
 
 /**
  * Converts an array or Tensor of integers to an int64 Tensor.
- * @param {Array|Tensor} items The input integers to be converted.
+ * @param {any[]|Tensor} items The input integers to be converted.
  * @returns {Tensor} The int64 Tensor with the converted values.
  * @throws {Error} If the input array is empty or the input is a batched Tensor and not all sequences have the same length.
  * @private
@@ -393,37 +444,6 @@ function toI64Tensor(items) {
     }
 }
 
-/**
- * Prepares an attention mask for a sequence of tokens based on configuration options.
- * @param {Object} self The calling object instance.
- * @param {Tensor} tokens The input tokens.
- * @returns {Tensor} The attention mask tensor.
- * @private
- */
-function prepareAttentionMask(self, tokens) {
-
-    // Prepare attention mask
-    let pad_token_id = self.config.pad_token_id ?? null;
-    let eos_token_id = self.config.eos_token_id ?? null;
-    if (isIntegralNumber(eos_token_id)) {
-        eos_token_id = [eos_token_id];
-    }
-
-    let is_pad_token_in_inputs = tokens.indexOf(pad_token_id) !== -1;
-    let is_pad_token_not_equal_to_eos_token_id = (eos_token_id === null) || !eos_token_id.includes(pad_token_id)
-
-    if (is_pad_token_in_inputs && is_pad_token_not_equal_to_eos_token_id) {
-        let data = BigInt64Array.from(
-            // Note: != so that int matches bigint
-            // @ts-ignore
-            tokens.data.map(x => x != pad_token_id)
-        )
-        return new Tensor('int64', data, tokens.dims)
-    } else {
-        return ones_like(tokens);
-    }
-}
-
 /**
  * Creates a boolean tensor with a single value.
  * @param {boolean} value The value of the tensor.
@@ -694,8 +714,8 @@ function image_text_to_text_prepare_inputs_for_generation(self, ...args) {
     } else {
         return decoder_prepare_inputs_for_generation(self, ...args);
     }
-
 }
+
 //////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
@@ -709,12 +729,14 @@ export class PreTrainedModel extends Callable {
      * Creates a new instance of the `PreTrainedModel` class.
      * @param {import('./configs.js').PretrainedConfig} config The model configuration.
      * @param {Record<string, any>} sessions The inference sessions for the model.
+     * @param {Record<string, Object>} configs Additional configuration files (e.g., generation_config.json).
      */
-    constructor(config, sessions) {
+    constructor(config, sessions, configs) {
         super();
 
         this.config = config;
         this.sessions = sessions;
+        this.configs = configs;
 
         const modelName = MODEL_CLASS_TO_NAME_MAPPING.get(this.constructor);
         const modelType = MODEL_TYPE_MAPPING.get(modelName);
@@ -830,7 +852,9 @@ export class PreTrainedModel extends Callable {
                 constructSessions(pretrained_model_name_or_path, {
                     model: options.model_file_name ?? 'model',
                 }, options),
-                getModelJSON(pretrained_model_name_or_path, 'generation_config.json', false, options),
+                getOptionalConfigs(pretrained_model_name_or_path, {
+                    generation_config: 'generation_config.json',
+                }, options),
             ]);
 
         } else if (modelType === MODEL_TYPES.Seq2Seq || modelType === MODEL_TYPES.Vision2Seq) {
@@ -839,7 +863,9 @@ export class PreTrainedModel extends Callable {
                     model: 'encoder_model',
                     decoder_model_merged: 'decoder_model_merged',
                 }, options),
-                getModelJSON(pretrained_model_name_or_path, 'generation_config.json', false, options),
+                getOptionalConfigs(pretrained_model_name_or_path, {
+                    generation_config: 'generation_config.json',
+                }, options),
             ]);
 
         } else if (modelType === MODEL_TYPES.MaskGeneration) {
@@ -869,7 +895,9 @@ export class PreTrainedModel extends Callable {
             }
             info = await Promise.all([
                 constructSessions(pretrained_model_name_or_path, sessions, options),
-                getModelJSON(pretrained_model_name_or_path, 'generation_config.json', false, options),
+                getOptionalConfigs(pretrained_model_name_or_path, {
+                    generation_config: 'generation_config.json',
+                }, options),
             ]);
 
         } else if (modelType === MODEL_TYPES.Musicgen) {
@@ -879,12 +907,14 @@ export class PreTrainedModel extends Callable {
                     decoder_model_merged: 'decoder_model_merged',
                     encodec_decode: 'encodec_decode',
                 }, options),
-                getModelJSON(pretrained_model_name_or_path, 'generation_config.json', false, options),
+                getOptionalConfigs(pretrained_model_name_or_path, {
+                    generation_config: 'generation_config.json',
+                }, options),
             ]);
 
         } else { // should be MODEL_TYPES.EncoderOnly
             if (modelType !== MODEL_TYPES.EncoderOnly) {
-                console.warn(`Model type for '${modelName ?? config?.model_type}' not found, assuming encoder-only architecture. Please report this at https://github.com/xenova/transformers.js/issues/new/choose.`)
+                console.warn(`Model type for '${modelName ?? config?.model_type}' not found, assuming encoder-only architecture. Please report this at ${GITHUB_ISSUE_URL}.`)
             }
             info = await Promise.all([
                 constructSessions(pretrained_model_name_or_path, {
@@ -917,6 +947,14 @@ export class PreTrainedModel extends Callable {
         return await this._forward(this, model_inputs);
     }
 
+    /**
+     * Get the model's generation config, if it exists.
+     * @returns {GenerationConfig|null} The model's generation config if it exists, otherwise `null`.
+     */
+    get generation_config() {
+        return this.configs?.generation_config ?? null;
+    }
+
     /**
      * This function returns a [`LogitsProcessorList`] list object that contains all relevant [`LogitsWarper`]
      * instances used for multinomial sampling.
@@ -1096,9 +1134,7 @@ export class PreTrainedModel extends Callable {
         const gen_config = new cls(config);
 
         // Apply model's generation config, if it exists
-        if ('generation_config' in this) {
-            Object.assign(gen_config, this.generation_config);
-        }
+        Object.assign(gen_config, this.generation_config ?? {});
 
         // Next, use any generation config specified by the user
         // when calling `generate`
@@ -1298,35 +1334,37 @@ export class PreTrainedModel extends Callable {
         let { decoder_input_ids, ...model_inputs } = model_kwargs;
 
         // Prepare input ids if the user has not defined `decoder_input_ids` manually.
-        if (!decoder_input_ids) {
-            decoder_start_token_id ??= bos_token_id;
-
-            if (this.config.model_type === 'musicgen') {
-                // Custom logic (TODO: move to Musicgen class)
-                decoder_input_ids = Array.from({
-                    length: batch_size * this.config.decoder.num_codebooks
-                }, () => [decoder_start_token_id]);
-
-            } else if (Array.isArray(decoder_start_token_id)) {
-                if (decoder_start_token_id.length !== batch_size) {
-                    throw new Error(
-                        `\`decoder_start_token_id\` expcted to have length ${batch_size} but got ${decoder_start_token_id.length}`
-                    )
+        if (!(decoder_input_ids instanceof Tensor)) {
+            if (!decoder_input_ids) {
+                decoder_start_token_id ??= bos_token_id;
+
+                if (this.config.model_type === 'musicgen') {
+                    // Custom logic (TODO: move to Musicgen class)
+                    decoder_input_ids = Array.from({
+                        length: batch_size * this.config.decoder.num_codebooks
+                    }, () => [decoder_start_token_id]);
+
+                } else if (Array.isArray(decoder_start_token_id)) {
+                    if (decoder_start_token_id.length !== batch_size) {
+                        throw new Error(
+                            `\`decoder_start_token_id\` expcted to have length ${batch_size} but got ${decoder_start_token_id.length}`
+                        )
+                    }
+                    decoder_input_ids = decoder_start_token_id;
+                } else {
+                    decoder_input_ids = Array.from({
+                        length: batch_size,
+                    }, () => [decoder_start_token_id]);
                 }
-                decoder_input_ids = decoder_start_token_id;
-            } else {
+            } else if (!Array.isArray(decoder_input_ids[0])) {
+                // Correct batch size
                 decoder_input_ids = Array.from({
                     length: batch_size,
-                }, () => [decoder_start_token_id]);
+                }, () => decoder_input_ids);
             }
-        } else if (!Array.isArray(decoder_input_ids[0])) {
-            // Correct batch size
-            decoder_input_ids = Array.from({
-                length: batch_size,
-            }, () => decoder_input_ids);
+            decoder_input_ids = toI64Tensor(decoder_input_ids);
         }
 
-        decoder_input_ids = toI64Tensor(decoder_input_ids);
         model_kwargs['decoder_attention_mask'] = ones_like(decoder_input_ids);
 
         return { input_ids: decoder_input_ids, model_inputs };
@@ -1458,13 +1496,12 @@ export class PreTrainedModel extends Callable {
         // - GenerationMode.BEAM_SEARCH
         // - GenerationMode.BEAM_SAMPLE
         ////////////////////////////////////////////////////
-        let past_key_values = null;
+        let outputs;
         let attentions = {};
         while (true) {
             // prepare model inputs
             model_inputs = this.prepare_inputs_for_generation(all_input_ids, model_inputs, generation_config);
-
-            const outputs = await this.forward(model_inputs);
+            outputs = await this.forward(model_inputs);
 
             if (generation_config.output_attentions && generation_config.return_dict_in_generate) {
                 // Get attentions if they are present
@@ -1511,10 +1548,6 @@ export class PreTrainedModel extends Callable {
 
             const stop = prepared_stopping_criteria(all_input_ids);
             if (stop.every(x => x)) {
-                if (generation_config.return_dict_in_generate) {
-                    // Get past key values without disposing buffers
-                    past_key_values = this.getPastKeyValues(outputs, model_inputs.past_key_values, false);
-                }
                 break;
             }
 
@@ -1527,6 +1560,9 @@ export class PreTrainedModel extends Callable {
             streamer.end();
         }
 
+        // Retrieve and dispose all final past key values (including encoder attentions)
+        const past_key_values = this.getPastKeyValues(outputs, model_inputs.past_key_values, true);
+
         // TODO: ensure all_input_ids is padded correctly...
         const sequences = new Tensor('int64', all_input_ids.flat(), [all_input_ids.length, all_input_ids[0].length]);
 
@@ -1540,6 +1576,12 @@ export class PreTrainedModel extends Callable {
                 // logits,
             }
         } else {
+            // Dispose all remaining tensors
+            for (const tensor of Object.values(outputs)) {
+                if (tensor.location === 'gpu-buffer') {
+                    tensor.dispose();
+                }
+            }
             return sequences;
         }
     }
@@ -1549,31 +1591,32 @@ export class PreTrainedModel extends Callable {
      *
      * @param {Object} decoderResults The decoder results object.
      * @param {Object} pastKeyValues The previous past key values.
-     * @param {boolean} [dispose=true] Whether to dispose of the old gpu buffer.
      * @returns {Object} An object containing past key values.
      */
-    getPastKeyValues(decoderResults, pastKeyValues, dispose = true) {
+    getPastKeyValues(decoderResults, pastKeyValues, disposeEncoderPKVs = false) {
         const pkvs = Object.create(null);
 
         for (const name in decoderResults) {
             if (name.startsWith('present')) {
                 const newName = name.replace('present', 'past_key_values');
-
-                if (pastKeyValues && name.includes('encoder')) {
-                    // Optimization introduced by optimum to reuse past key values. So, we just replace the constant
-                    // outputs with the previous past key values.
+                const is_encoder_pkv = name.includes('encoder');
+                if (is_encoder_pkv && pastKeyValues) {
+                    // Optimization introduced by optimum to reuse past key values.
+                    // So, we just replace the constant outputs (`decoderResults[name]`) with the previous past key values.
                     // https://github.com/huggingface/optimum/blob/0bf2c05fb7e1182b52d21b703cfc95fd9e4ea3dc/optimum/onnxruntime/base.py#L677-L704
                     pkvs[newName] = pastKeyValues[newName];
-                } else {
-                    if (dispose && pastKeyValues) {
-                        // Free old gpu buffer
-                        const t = pastKeyValues[newName];
-                        if (t.location === 'gpu-buffer') {
-                            t.dispose();
-                        }
-                    }
+                } else { // decoder or using first encoder PKVs
                     pkvs[newName] = decoderResults[name];
                 }
+
+                if (pastKeyValues && (!is_encoder_pkv || disposeEncoderPKVs)) {
+                    // - Always dispose decoder PKVs
+                    // - Only dispose encoder past key values when requested (after generation)
+                    const t = pastKeyValues[newName];
+                    if (t.location === 'gpu-buffer') {
+                        t.dispose();
+                    }
+                }
             }
         }
         return pkvs;
@@ -1611,9 +1654,8 @@ export class PreTrainedModel extends Callable {
         if (pastKeyValues) {
             Object.assign(decoderFeeds, pastKeyValues)
         } else {
-
-            /** @type {import('./transformers.js').DataType} */
-            const dtype = this.custom_config.kv_cache_dtype ?? 'float32';
+            const session = this.sessions['decoder_model_merged'] ?? this.sessions['model'];
+            const dtype = session?.config?.kv_cache_dtype ?? 'float32';
             const empty = (dtype === 'float16') ? new Uint16Array() : [];
 
             const shapes = getKeyValueShapes(this.config);
@@ -2506,17 +2548,6 @@ export class T5PreTrainedModel extends PreTrainedModel {
         'decoder_attention_mask',
         'past_key_values',
     ];
-
-    /**
-     * Creates a new instance of the `T5PreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
 };
 
 export class T5Model extends T5PreTrainedModel { }
@@ -2533,18 +2564,7 @@ export class T5ForConditionalGeneration extends T5PreTrainedModel { }
 /**
  * An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.
  */
-export class LongT5PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `LongT5ForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class LongT5PreTrainedModel extends PreTrainedModel { };
 
 /**
  * The bare LONGT5 Model transformer outputting raw hidden-states without any specific head on top.
@@ -2560,19 +2580,7 @@ export class LongT5ForConditionalGeneration extends LongT5PreTrainedModel { }
 
 //////////////////////////////////////////////////
 // MT5 models
-export class MT5PreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `MT5ForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class MT5PreTrainedModel extends PreTrainedModel { };
 
 export class MT5Model extends MT5PreTrainedModel { }
 
@@ -2584,19 +2592,7 @@ export class MT5ForConditionalGeneration extends MT5PreTrainedModel { }
 
 //////////////////////////////////////////////////
 // Bart models
-export class BartPretrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `BartForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class BartPretrainedModel extends PreTrainedModel { };
 
 /**
  * The bare BART Model outputting raw hidden-states without any specific head on top.
@@ -2627,19 +2623,7 @@ export class BartForSequenceClassification extends BartPretrainedModel {
 
 //////////////////////////////////////////////////
 // MBart models
-export class MBartPreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `MBartForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class MBartPreTrainedModel extends PreTrainedModel { };
 
 /**
  * The bare MBART Model outputting raw hidden-states without any specific head on top.
@@ -2673,19 +2657,7 @@ export class MBartForCausalLM extends MBartPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // Blenderbot models
-export class BlenderbotPreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `BlenderbotForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class BlenderbotPreTrainedModel extends PreTrainedModel { };
 
 /**
  * The bare Blenderbot Model outputting raw hidden-states without any specific head on top.
@@ -2701,19 +2673,7 @@ export class BlenderbotForConditionalGeneration extends BlenderbotPreTrainedMode
 
 //////////////////////////////////////////////////
 // Blenderbot models
-export class BlenderbotSmallPreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `BlenderbotForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class BlenderbotSmallPreTrainedModel extends PreTrainedModel { };
 
 /**
  * The bare BlenderbotSmall Model outputting raw hidden-states without any specific head on top.
@@ -2962,17 +2922,6 @@ export class WhisperPreTrainedModel extends PreTrainedModel {
         'decoder_attention_mask',
         'past_key_values',
     ];
-
-    /**
-     * Creates a new instance of the `WhisperPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
 };
 
 /**
@@ -3238,21 +3187,14 @@ export class WhisperForConditionalGeneration extends WhisperPreTrainedModel {
 export class VisionEncoderDecoderModel extends PreTrainedModel {
     main_input_name = 'pixel_values';
     forward_params = [
+        // Encoder inputs
         'pixel_values',
-        'input_ids',
+
+        // Decoder inpputs
+        'decoder_input_ids',
         'encoder_hidden_states',
         'past_key_values',
     ];
-    /**
-     * Creates a new instance of the `VisionEncoderDecoderModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
 }
 //////////////////////////////////////////////////
 
@@ -3267,11 +3209,6 @@ export class LlavaPreTrainedModel extends PreTrainedModel {
         'position_ids',
         'past_key_values',
     ];
-
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
 }
 
 /**
@@ -3358,11 +3295,6 @@ export class Florence2PreTrainedModel extends PreTrainedModel {
         'past_key_values',
     ];
     main_input_name = 'inputs_embeds';
-
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
 }
 
 export class Florence2ForConditionalGeneration extends Florence2PreTrainedModel {
@@ -3501,6 +3433,18 @@ export class CLIPPreTrainedModel extends PreTrainedModel { }
  */
 export class CLIPModel extends CLIPPreTrainedModel { }
 
+/**
+ * The text model from CLIP without any head or projection on top.
+ */
+export 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)
  * 
@@ -3528,7 +3472,6 @@ export class CLIPModel extends CLIPPreTrainedModel { }
  * ```
  */
 export 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
@@ -3537,6 +3480,18 @@ export class CLIPTextModelWithProjection extends CLIPPreTrainedModel {
     }
 }
 
+/**
+ * The vision model from CLIP without any head or projection on top.
+ */
+export 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)
  * 
@@ -3759,18 +3714,7 @@ export class CLIPSegForImageSegmentation extends CLIPSegPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // GPT2 models
-export class GPT2PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GPT2PreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GPT2PreTrainedModel extends PreTrainedModel { }
 
 export class GPT2Model extends GPT2PreTrainedModel { }
 
@@ -3783,20 +3727,25 @@ export class GPT2LMHeadModel extends GPT2PreTrainedModel { }
 // }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+// JAIS models
+export class JAISPreTrainedModel extends PreTrainedModel { }
+
+/**
+ * The bare JAIS Model transformer outputting raw hidden-states without any specific head on top.
+ */
+export class JAISModel extends JAISPreTrainedModel { }
+
+/**
+ * The JAIS Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).
+ */
+export class JAISLMHeadModel extends JAISPreTrainedModel { }
+//////////////////////////////////////////////////
+
+
 //////////////////////////////////////////////////
 // GPTNeo models
-export class GPTNeoPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GPTNeoPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GPTNeoPreTrainedModel extends PreTrainedModel { }
 export class GPTNeoModel extends GPTNeoPreTrainedModel { }
 
 export class GPTNeoForCausalLM extends GPTNeoPreTrainedModel { }
@@ -3804,18 +3753,7 @@ export class GPTNeoForCausalLM extends GPTNeoPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // GPTNeoX models
-export class GPTNeoXPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GPTNeoXPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GPTNeoXPreTrainedModel extends PreTrainedModel { }
 export class GPTNeoXModel extends GPTNeoXPreTrainedModel { }
 
 export class GPTNeoXForCausalLM extends GPTNeoXPreTrainedModel { }
@@ -3824,18 +3762,7 @@ export class GPTNeoXForCausalLM extends GPTNeoXPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // GPT-J models
-export class GPTJPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GPTJPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GPTJPreTrainedModel extends PreTrainedModel { }
 
 export class GPTJModel extends GPTJPreTrainedModel { }
 
@@ -3845,18 +3772,7 @@ export class GPTJForCausalLM extends GPTJPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // GPTBigCode models
-export class GPTBigCodePreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GPTBigCodePreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GPTBigCodePreTrainedModel extends PreTrainedModel { }
 
 export class GPTBigCodeModel extends GPTBigCodePreTrainedModel { }
 
@@ -3865,18 +3781,7 @@ export class GPTBigCodeForCausalLM extends GPTBigCodePreTrainedModel { }
 
 //////////////////////////////////////////////////
 // CodeGen models
-export class CodeGenPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `CodeGenPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class CodeGenPreTrainedModel extends PreTrainedModel { }
 /**
  * CodeGenModel is a class representing a code generation model without a language model head.
  */
@@ -3895,18 +3800,7 @@ export class CodeGenForCausalLM extends CodeGenPreTrainedModel { }
 /**
  * The bare LLama Model outputting raw hidden-states without any specific head on top.
  */
-export class LlamaPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `LlamaPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class LlamaPreTrainedModel extends PreTrainedModel { }
 /**
  * The bare LLaMA Model outputting raw hidden-states without any specific head on top.
  */
@@ -3915,24 +3809,22 @@ export class LlamaModel extends LlamaPreTrainedModel { }
 export class LlamaForCausalLM extends LlamaPreTrainedModel { }
 //////////////////////////////////////////////////
 
+
+//////////////////////////////////////////////////
+// Granite models
+export class GranitePreTrainedModel extends PreTrainedModel { }
+export class GraniteModel extends GranitePreTrainedModel { }
+export class GraniteForCausalLM extends GranitePreTrainedModel { }
+//////////////////////////////////////////////////
+
+
 //////////////////////////////////////////////////
 // Cohere models
 
 /**
  * The bare Cohere Model outputting raw hidden-states without any specific head on top.
  */
-export class CoherePreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `CoherePreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class CoherePreTrainedModel extends PreTrainedModel { }
 export class CohereModel extends CoherePreTrainedModel { }
 
 export class CohereForCausalLM extends CoherePreTrainedModel { }
@@ -3944,18 +3836,7 @@ export class CohereForCausalLM extends CoherePreTrainedModel { }
 /**
  * The bare Gemma Model outputting raw hidden-states without any specific head on top.
  */
-export class GemmaPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `GemmaPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class GemmaPreTrainedModel extends PreTrainedModel { }
 /**
  * The bare Gemma Model outputting raw hidden-states without any specific head on top.
  */
@@ -3970,18 +3851,7 @@ export class GemmaForCausalLM extends GemmaPreTrainedModel { }
 /**
  * The bare Gemma2 Model outputting raw hidden-states without any specific head on top.
  */
-export class Gemma2PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `Gemma2PreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class Gemma2PreTrainedModel extends PreTrainedModel { }
 /**
  * The bare Gemma2 Model outputting raw hidden-states without any specific head on top.
  */
@@ -3991,18 +3861,7 @@ export class Gemma2ForCausalLM extends Gemma2PreTrainedModel { }
 //////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
-export class OpenELMPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `OpenELMPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class OpenELMPreTrainedModel extends PreTrainedModel { }
 export class OpenELMModel extends OpenELMPreTrainedModel { }
 
 export class OpenELMForCausalLM extends OpenELMPreTrainedModel { }
@@ -4014,18 +3873,7 @@ export class OpenELMForCausalLM extends OpenELMPreTrainedModel { }
 /**
  * The bare Qwen2 Model outputting raw hidden-states without any specific head on top.
  */
-export class Qwen2PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `Qwen2PreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class Qwen2PreTrainedModel extends PreTrainedModel { }
 /**
  * The bare Qwen2 Model outputting raw hidden-states without any specific head on top.
  */
@@ -4037,18 +3885,7 @@ export class Qwen2ForCausalLM extends Qwen2PreTrainedModel { }
 
 //////////////////////////////////////////////////
 // Phi models
-export class PhiPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `PhiPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class PhiPreTrainedModel extends PreTrainedModel { }
 /**
  * The bare Phi Model outputting raw hidden-states without any specific head on top.
  */
@@ -4059,18 +3896,7 @@ export class PhiForCausalLM extends PhiPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // Phi3 models
-export class Phi3PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `Phi3PreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class Phi3PreTrainedModel extends PreTrainedModel { }
 
 /**
  * The bare Phi3 Model outputting raw hidden-states without any specific head on top.
@@ -4086,18 +3912,7 @@ export class Phi3ForCausalLM extends Phi3PreTrainedModel { }
 /**
  * The Bloom Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).
  */
-export class BloomPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `BloomPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class BloomPreTrainedModel extends PreTrainedModel { }
 
 /**
  * The bare Bloom Model transformer outputting raw hidden-states without any specific head on top.
@@ -4112,18 +3927,7 @@ export class BloomForCausalLM extends BloomPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // MPT models
-export class MptPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `MptPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class MptPreTrainedModel extends PreTrainedModel { }
 
 /**
  * The bare Mpt Model transformer outputting raw hidden-states without any specific head on top.
@@ -4139,18 +3943,7 @@ export class MptForCausalLM extends MptPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // OPT models
-export class OPTPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `OPTPreTrainedModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-}
+export class OPTPreTrainedModel extends PreTrainedModel { }
 
 /**
  * The bare OPT Model outputting raw hidden-states without any specific head on top.
@@ -4176,6 +3969,43 @@ export class ViTForImageClassification extends ViTPreTrainedModel {
 }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+export class PvtPreTrainedModel extends PreTrainedModel { }
+export class PvtModel extends PvtPreTrainedModel { }
+export class PvtForImageClassification extends PvtPreTrainedModel {
+    /**
+     * @param {any} model_inputs
+     */
+    async _call(model_inputs) {
+        return new SequenceClassifierOutput(await super._call(model_inputs));
+    }
+}
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+export class ViTMAEPreTrainedModel extends PreTrainedModel { }
+export class ViTMAEModel extends ViTMAEPreTrainedModel { }
+//////////////////////////////////////////////////
+
+
+//////////////////////////////////////////////////
+export class ViTMSNPreTrainedModel extends PreTrainedModel { }
+export class ViTMSNModel extends ViTMSNPreTrainedModel { }
+export class ViTMSNForImageClassification extends ViTMSNPreTrainedModel {
+    /**
+     * @param {any} model_inputs
+     */
+    async _call(model_inputs) {
+        return new SequenceClassifierOutput(await super._call(model_inputs));
+    }
+}
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+export class GroupViTPreTrainedModel extends PreTrainedModel { }
+export class GroupViTModel extends GroupViTPreTrainedModel { }
+//////////////////////////////////////////////////
+
 
 //////////////////////////////////////////////////
 export class FastViTPreTrainedModel extends PreTrainedModel { }
@@ -4429,6 +4259,19 @@ export class DeiTForImageClassification extends DeiTPreTrainedModel {
 }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+export class HieraPreTrainedModel extends PreTrainedModel { }
+export class HieraModel extends HieraPreTrainedModel { }
+export class HieraForImageClassification extends HieraPreTrainedModel {
+    /**
+     * @param {any} model_inputs
+     */
+    async _call(model_inputs) {
+        return new SequenceClassifierOutput(await super._call(model_inputs));
+    }
+}
+//////////////////////////////////////////////////
+
 
 //////////////////////////////////////////////////
 /**
@@ -4568,6 +4411,24 @@ export class DepthAnythingForDepthEstimation extends DepthAnythingPreTrainedMode
 //////////////////////////////////////////////////
 
 
+//////////////////////////////////////////////////
+export class SapiensPreTrainedModel extends PreTrainedModel { }
+export class SapiensForSemanticSegmentation extends SapiensPreTrainedModel { }
+export class SapiensForDepthEstimation extends SapiensPreTrainedModel { }
+export class SapiensForNormalEstimation extends SapiensPreTrainedModel { }
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+export class DepthProPreTrainedModel extends PreTrainedModel { }
+export class DepthProForDepthEstimation extends DepthProPreTrainedModel { }
+//////////////////////////////////////////////////
+
+//////////////////////////////////////////////////
+export class MaskFormerPreTrainedModel extends PreTrainedModel { }
+export class MaskFormerModel extends MaskFormerPreTrainedModel { }
+export class MaskFormerForInstanceSegmentation extends MaskFormerPreTrainedModel { }
+//////////////////////////////////////////////////
+
 //////////////////////////////////////////////////
 export class GLPNPreTrainedModel extends PreTrainedModel { }
 
@@ -4944,19 +4805,7 @@ export class SamImageSegmentationOutput extends ModelOutput {
 
 //////////////////////////////////////////////////
 // MarianMT models
-export class MarianPreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `MarianMTModel` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class MarianPreTrainedModel extends PreTrainedModel { };
 
 export class MarianModel extends MarianPreTrainedModel { }
 
@@ -4965,19 +4814,7 @@ export class MarianMTModel extends MarianPreTrainedModel { }
 
 //////////////////////////////////////////////////
 // M2M100 models
-export class M2M100PreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `M2M100ForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class M2M100PreTrainedModel extends PreTrainedModel { };
 
 export class M2M100Model extends M2M100PreTrainedModel { }
 
@@ -5069,7 +4906,7 @@ export class PyAnnoteModel extends PyAnnotePreTrainedModel { }
  * **Example:** Load and run a `PyAnnoteForAudioFrameClassification` for speaker diarization.
  * 
  * ```javascript
- * import { AutoProcessor, AutoModelForAudioFrameClassification, read_audio } from '@xenova/transformers';
+ * import { AutoProcessor, AutoModelForAudioFrameClassification, read_audio } from '@huggingface/transformers';
  * 
  * // Load model and processor
  * const model_id = 'onnx-community/pyannote-segmentation-3.0';
@@ -5487,19 +5324,7 @@ export class WavLMForAudioFrameClassification extends WavLMPreTrainedModel {
 /**
  * An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.
  */
-export class SpeechT5PreTrainedModel extends PreTrainedModel {
-
-    /**
-     * Creates a new instance of the `SpeechT5ForTextToSpeech` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-};
+export class SpeechT5PreTrainedModel extends PreTrainedModel { };
 
 /**
  * The bare SpeechT5 Encoder-Decoder Model outputting raw hidden-states without any specific pre- or post-nets.
@@ -5660,18 +5485,7 @@ export class SpeechT5HifiGan extends PreTrainedModel {
 
 //////////////////////////////////////////////////
 // TrOCR models
-export class TrOCRPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `TrOCRPreTrainedModel` class.
-     * @param {Object} config The configuration of the model.
-     * @param {any} session The ONNX session containing the model weights.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, session, generation_config) {
-        super(config, session);
-        this.generation_config = generation_config;
-    }
-}
+export class TrOCRPreTrainedModel extends PreTrainedModel { }
 
 /**
  * The TrOCR Decoder with a language modeling head.
@@ -5686,18 +5500,7 @@ export class TrOCRForCausalLM extends TrOCRPreTrainedModel { }
 /**
  * The bare Mistral Model outputting raw hidden-states without any specific head on top.
  */
-export class MistralPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `MistralPreTrainedModel` class.
-     * @param {Object} config The configuration of the model.
-     * @param {any} session The ONNX session containing the model weights.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, session, generation_config) {
-        super(config, session);
-        this.generation_config = generation_config;
-    }
-}
+export class MistralPreTrainedModel extends PreTrainedModel { }
 
 export class MistralModel extends MistralPreTrainedModel { }
 
@@ -5710,18 +5513,7 @@ export class MistralForCausalLM extends MistralPreTrainedModel { }
 /**
  * The bare Starcoder2 Model outputting raw hidden-states without any specific head on top.
  */
-export class Starcoder2PreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `Starcoder2PreTrainedModel` class.
-     * @param {Object} config The configuration of the model.
-     * @param {any} session The ONNX session containing the model weights.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, session, generation_config) {
-        super(config, session);
-        this.generation_config = generation_config;
-    }
-}
+export class Starcoder2PreTrainedModel extends PreTrainedModel { }
 
 export class Starcoder2Model extends Starcoder2PreTrainedModel { }
 
@@ -5734,18 +5526,7 @@ export class Starcoder2ForCausalLM extends Starcoder2PreTrainedModel { }
 /**
  * The bare Falcon Model outputting raw hidden-states without any specific head on top.
  */
-export class FalconPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `FalconPreTrainedModel` class.
-     * @param {Object} config The configuration of the model.
-     * @param {any} session The ONNX session containing the model weights.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, session, generation_config) {
-        super(config, session);
-        this.generation_config = generation_config;
-    }
-}
+export class FalconPreTrainedModel extends PreTrainedModel { }
 
 export class FalconModel extends FalconPreTrainedModel { }
 
@@ -5895,18 +5676,7 @@ export class SegformerForSemanticSegmentation extends SegformerPreTrainedModel {
 
 //////////////////////////////////////////////////
 // StableLm models
-export class StableLmPreTrainedModel extends PreTrainedModel {
-    /**
-     * Creates a new instance of the `StableLmPreTrainedModel` class.
-     * @param {Object} config The configuration of the model.
-     * @param {any} session The ONNX session containing the model weights.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, session, generation_config) {
-        super(config, session);
-        this.generation_config = generation_config;
-    }
-}
+export class StableLmPreTrainedModel extends PreTrainedModel { }
 
 /**
  * The bare StableLm Model transformer outputting raw hidden-states without any specific head on top.
@@ -6000,17 +5770,6 @@ export class MusicgenForConditionalGeneration extends PreTrainedModel { // NOTE:
         'past_key_values',
     ];
 
-    /**
-     * Creates a new instance of the `MusicgenForConditionalGeneration` class.
-     * @param {Object} config The model configuration.
-     * @param {Record<string, any>} sessions The inference sessions for the model.
-     * @param {GenerationConfig} generation_config The generation configuration.
-     */
-    constructor(config, sessions, generation_config) {
-        super(config, sessions);
-        this.generation_config = generation_config;
-    }
-
     /**
      * Apply the pattern mask to the final ids,
      * then revert the pattern delay mask by filtering the pad token id in a single step.
@@ -6089,6 +5848,7 @@ export class MusicgenForConditionalGeneration extends PreTrainedModel { // NOTE:
         return audio_values;
     }
 }
+//////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
 // MobileNetV1 models
@@ -6182,6 +5942,17 @@ export class MobileNetV4ForImageClassification extends MobileNetV4PreTrainedMode
 }
 //////////////////////////////////////////////////
 
+//////////////////////////////////////////////////
+// Decision Transformer models
+export 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
+ */
+export class DecisionTransformerModel extends DecisionTransformerPreTrainedModel { }
+
+//////////////////////////////////////////////////
 
 //////////////////////////////////////////////////
 // AutoModels, used to simplify construction of PreTrainedModels
@@ -6220,7 +5991,7 @@ export class PretrainedMixin {
         session_options = {},
     } = {}) {
 
-        let options = {
+        const options = {
             progress_callback,
             config,
             cache_dir,
@@ -6239,7 +6010,7 @@ export 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
@@ -6294,6 +6065,10 @@ 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]],
@@ -6301,6 +6076,7 @@ const MODEL_MAPPING_NAMES_ENCODER_ONLY = new Map([
     ['owlv2', ['Owlv2Model', Owlv2Model]],
     ['beit', ['BeitModel', BeitModel]],
     ['deit', ['DeiTModel', DeiTModel]],
+    ['hiera', ['HieraModel', HieraModel]],
     ['convnext', ['ConvNextModel', ConvNextModel]],
     ['convnextv2', ['ConvNextV2Model', ConvNextV2Model]],
     ['dinov2', ['Dinov2Model', Dinov2Model]],
@@ -6315,10 +6091,14 @@ 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([
@@ -6337,6 +6117,7 @@ const MODEL_MAPPING_NAMES_ENCODER_DECODER = new Map([
 
 const MODEL_MAPPING_NAMES_DECODER_ONLY = new Map([
     ['bloom', ['BloomModel', BloomModel]],
+    ['jais', ['JAISModel', JAISModel]],
     ['gpt2', ['GPT2Model', GPT2Model]],
     ['gptj', ['GPTJModel', GPTJModel]],
     ['gpt_bigcode', ['GPTBigCodeModel', GPTBigCodeModel]],
@@ -6344,6 +6125,7 @@ const MODEL_MAPPING_NAMES_DECODER_ONLY = new Map([
     ['gpt_neox', ['GPTNeoXModel', GPTNeoXModel]],
     ['codegen', ['CodeGenModel', CodeGenModel]],
     ['llama', ['LlamaModel', LlamaModel]],
+    ['granite', ['GraniteModel', GraniteModel]],
     ['cohere', ['CohereModel', CohereModel]],
     ['gemma', ['GemmaModel', GemmaModel]],
     ['gemma2', ['Gemma2Model', Gemma2Model]],
@@ -6425,12 +6207,14 @@ const MODEL_FOR_SEQ_TO_SEQ_CAUSAL_LM_MAPPING_NAMES = new Map([
 const MODEL_FOR_CAUSAL_LM_MAPPING_NAMES = new Map([
     ['bloom', ['BloomForCausalLM', BloomForCausalLM]],
     ['gpt2', ['GPT2LMHeadModel', GPT2LMHeadModel]],
+    ['jais', ['JAISLMHeadModel', JAISLMHeadModel]],
     ['gptj', ['GPTJForCausalLM', GPTJForCausalLM]],
     ['gpt_bigcode', ['GPTBigCodeForCausalLM', GPTBigCodeForCausalLM]],
     ['gpt_neo', ['GPTNeoForCausalLM', GPTNeoForCausalLM]],
     ['gpt_neox', ['GPTNeoXForCausalLM', GPTNeoXForCausalLM]],
     ['codegen', ['CodeGenForCausalLM', CodeGenForCausalLM]],
     ['llama', ['LlamaForCausalLM', LlamaForCausalLM]],
+    ['granite', ['GraniteForCausalLM', GraniteForCausalLM]],
     ['cohere', ['CohereForCausalLM', CohereForCausalLM]],
     ['gemma', ['GemmaForCausalLM', GemmaForCausalLM]],
     ['gemma2', ['Gemma2ForCausalLM', Gemma2ForCausalLM]],
@@ -6501,11 +6285,14 @@ 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]],
     ['beit', ['BeitForImageClassification', BeitForImageClassification]],
     ['deit', ['DeiTForImageClassification', DeiTForImageClassification]],
+    ['hiera', ['HieraForImageClassification', HieraForImageClassification]],
     ['convnext', ['ConvNextForImageClassification', ConvNextForImageClassification]],
     ['convnextv2', ['ConvNextV2ForImageClassification', ConvNextV2ForImageClassification]],
     ['dinov2', ['Dinov2ForImageClassification', Dinov2ForImageClassification]],
@@ -6532,12 +6319,19 @@ 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]],
 ]);
 
 const MODEL_FOR_SEMANTIC_SEGMENTATION_MAPPING_NAMES = new Map([
     ['segformer', ['SegformerForSemanticSegmentation', SegformerForSemanticSegmentation]],
+    ['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([
@@ -6586,6 +6380,12 @@ const MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES = new Map([
     ['dpt', ['DPTForDepthEstimation', DPTForDepthEstimation]],
     ['depth_anything', ['DepthAnythingForDepthEstimation', DepthAnythingForDepthEstimation]],
     ['glpn', ['GLPNForDepthEstimation', GLPNForDepthEstimation]],
+    ['sapiens', ['SapiensForDepthEstimation', SapiensForDepthEstimation]],
+    ['depth_pro', ['DepthProForDepthEstimation', DepthProForDepthEstimation]],
+])
+
+const MODEL_FOR_NORMAL_ESTIMATION_MAPPING_NAMES = new Map([
+    ['sapiens', ['SapiensForNormalEstimation', SapiensForNormalEstimation]],
 ])
 
 // NOTE: This is custom to Transformers.js, and is necessary because certain models
@@ -6610,10 +6410,12 @@ 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],
     [MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
+    [MODEL_FOR_NORMAL_ESTIMATION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_OBJECT_DETECTION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_ZERO_SHOT_OBJECT_DETECTION_MAPPING_NAMES, MODEL_TYPES.EncoderOnly],
     [MODEL_FOR_MASK_GENERATION_MAPPING_NAMES, MODEL_TYPES.MaskGeneration],
@@ -6811,6 +6613,17 @@ export 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');
+ */
+export 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.
@@ -6870,6 +6683,10 @@ export class AutoModelForDepthEstimation extends PretrainedMixin {
     static MODEL_CLASS_MAPPINGS = [MODEL_FOR_DEPTH_ESTIMATION_MAPPING_NAMES];
 }
 
+export class AutoModelForNormalEstimation extends PretrainedMixin {
+    static MODEL_CLASS_MAPPINGS = [MODEL_FOR_NORMAL_ESTIMATION_MAPPING_NAMES];
+}
+
 export class AutoModelForImageFeatureExtraction extends PretrainedMixin {
     static MODEL_CLASS_MAPPINGS = [MODEL_FOR_IMAGE_FEATURE_EXTRACTION_MAPPING_NAMES];
 }