models

Definitions of all models available in Transformers.js.

Example: Load and run an AutoModel.

import { AutoModel, AutoTokenizer } from '@huggingface/transformers';

let tokenizer = await AutoTokenizer.from_pretrained('Xenova/bert-base-uncased');
let model = await AutoModel.from_pretrained('Xenova/bert-base-uncased');

let inputs = await tokenizer('I love transformers!');
let { logits } = await model(inputs);
// Tensor {
//     data: Float32Array(183132) [-7.117443084716797, -7.107812881469727, -7.092104911804199, ...]
//     dims: (3) [1, 6, 30522],
//     type: "float32",
//     size: 183132,
// }

We also provide other AutoModels (listed below), which you can use in the same way as the Python library. For example:

Example: Load and run an AutoModelForSeq2SeqLM.

import { AutoModelForSeq2SeqLM, AutoTokenizer } from '@huggingface/transformers';

let tokenizer = await AutoTokenizer.from_pretrained('Xenova/t5-small');
let model = await AutoModelForSeq2SeqLM.from_pretrained('Xenova/t5-small');

let { input_ids } = await tokenizer('translate English to German: I love transformers!');
let outputs = await model.generate(input_ids);
let decoded = tokenizer.decode(outputs[0], { skip_special_tokens: true });
// 'Ich liebe Transformatoren!'

• models
  • static
    • .PreTrainedModel
      • new PreTrainedModel(config, sessions, configs)
      • instance
        • .custom_config : *
        • .generation_config ⇒ GenerationConfig | null
        • .dispose() ⇒ Promise.<Array<unknown>>
        • ._call(model_inputs) ⇒ Promise.<Object>
        • .forward(model_inputs) ⇒ Promise.<Object>
        • ._get_logits_warper(generation_config) ⇒ LogitsProcessorList
        • ._prepare_generation_config(generation_config, kwargs) ⇒ GenerationConfig
        • ._get_stopping_criteria(generation_config, [stopping_criteria])
        • ._validate_model_class()
        • ._update_model_kwargs_for_generation(inputs) ⇒ Object
        • ._prepare_model_inputs(params) ⇒ Object
        • ._prepare_decoder_input_ids_for_generation(param0)
        • .generate(options) ⇒ Promise.<(ModelOutput|Tensor)>
        • .getPastKeyValues(decoderResults, pastKeyValues) ⇒ Object
        • .getAttentions(model_output) ⇒ *
        • .addPastKeyValues(decoderFeeds, pastKeyValues)
      • static
        • .from_pretrained(pretrained_model_name_or_path, options) ⇒ Promise.<PreTrainedModel>
    • .BaseModelOutput
      • new BaseModelOutput(output)
    • .BertForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .BertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .BertForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .BertForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .RoFormerModel
    • .RoFormerForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .RoFormerForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .RoFormerForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .RoFormerForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .ConvBertModel
    • .ConvBertForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .ConvBertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .ConvBertForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .ConvBertForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .ElectraModel
    • .ElectraForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .ElectraForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .ElectraForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .ElectraForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .CamembertModel
    • .CamembertForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .CamembertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .CamembertForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .CamembertForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .DebertaModel
    • .DebertaForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .DebertaForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .DebertaForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .DebertaForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .DebertaV2Model
    • .DebertaV2ForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .DebertaV2ForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .DebertaV2ForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .DebertaV2ForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .DistilBertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .DistilBertForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .DistilBertForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .DistilBertForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .EsmModel
    • .EsmForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .EsmForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .EsmForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .MobileBertForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .MobileBertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .MobileBertForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .MPNetModel
    • .MPNetForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .MPNetForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .MPNetForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .MPNetForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .T5ForConditionalGeneration
    • .LongT5PreTrainedModel
    • .LongT5Model
    • .LongT5ForConditionalGeneration
    • .MT5ForConditionalGeneration
    • .BartModel
    • .BartForConditionalGeneration
    • .BartForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .MBartModel
    • .MBartForConditionalGeneration
    • .MBartForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .BlenderbotModel
    • .BlenderbotForConditionalGeneration
    • .BlenderbotSmallModel
    • .BlenderbotSmallForConditionalGeneration
    • .RobertaForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .RobertaForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .RobertaForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .RobertaForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .XLMPreTrainedModel
    • .XLMModel
    • .XLMWithLMHeadModel
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .XLMForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .XLMForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .XLMForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .XLMRobertaForMaskedLM
      • ._call(model_inputs) ⇒ Promise.<MaskedLMOutput>
    • .XLMRobertaForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .XLMRobertaForTokenClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .XLMRobertaForQuestionAnswering
      • ._call(model_inputs) ⇒ Promise.<QuestionAnsweringModelOutput>
    • .ASTModel
    • .ASTForAudioClassification
    • .WhisperModel
    • .WhisperForConditionalGeneration
      • ._retrieve_init_tokens(generation_config)
      • .generate(options) ⇒ Promise.<(ModelOutput|Tensor)>
      • ._extract_token_timestamps(generate_outputs, alignment_heads, [num_frames], [time_precision]) ⇒ Tensor
    • .MoonshineModel
    • .VisionEncoderDecoderModel
    • .LlavaForConditionalGeneration
    • .Idefics3ForConditionalGeneration
    • .CLIPModel
    • .CLIPTextModel
      • .from_pretrained() : *
    • .CLIPTextModelWithProjection
      • .from_pretrained() : *
    • .CLIPVisionModel
      • .from_pretrained() : *
    • .CLIPVisionModelWithProjection
      • .from_pretrained() : *
    • .SiglipModel
    • .SiglipTextModel
      • .from_pretrained() : *
    • .SiglipVisionModel
      • .from_pretrained() : *
    • .CLIPSegForImageSegmentation
    • .GPT2LMHeadModel
    • .JAISModel
    • .JAISLMHeadModel
    • .CodeGenModel
    • .CodeGenForCausalLM
    • .LlamaPreTrainedModel
    • .LlamaModel
    • .CoherePreTrainedModel
    • .GemmaPreTrainedModel
    • .GemmaModel
    • .Gemma2PreTrainedModel
    • .Gemma2Model
    • .Qwen2PreTrainedModel
    • .Qwen2Model
    • .PhiModel
    • .Phi3Model
    • .BloomPreTrainedModel
    • .BloomModel
    • .BloomForCausalLM
    • .MptModel
    • .MptForCausalLM
    • .OPTModel
    • .OPTForCausalLM
    • .VitPoseForPoseEstimation
    • .VitMatteForImageMatting
      • ._call(model_inputs)
    • .DetrObjectDetectionOutput
      • new DetrObjectDetectionOutput(output)
    • .DetrSegmentationOutput
      • new DetrSegmentationOutput(output)
    • .RTDetrObjectDetectionOutput
      • new RTDetrObjectDetectionOutput(output)
    • .TableTransformerModel
    • .TableTransformerForObjectDetection
      • ._call(model_inputs)
    • .ResNetPreTrainedModel
    • .ResNetModel
    • .ResNetForImageClassification
      • ._call(model_inputs)
    • .Swin2SRModel
    • .Swin2SRForImageSuperResolution
    • .DPTModel
    • .DPTForDepthEstimation
    • .DepthAnythingForDepthEstimation
    • .GLPNModel
    • .GLPNForDepthEstimation
    • .DonutSwinModel
    • .ConvNextModel
    • .ConvNextForImageClassification
      • ._call(model_inputs)
    • .ConvNextV2Model
    • .ConvNextV2ForImageClassification
      • ._call(model_inputs)
    • .Dinov2Model
    • .Dinov2ForImageClassification
      • ._call(model_inputs)
    • .Dinov2WithRegistersModel
    • .Dinov2WithRegistersForImageClassification
      • ._call(model_inputs)
    • .YolosObjectDetectionOutput
      • new YolosObjectDetectionOutput(output)
    • .SamModel
      • .get_image_embeddings(model_inputs) ⇒ Promise.<{image_embeddings: Tensor, image_positional_embeddings: Tensor}>
      • .forward(model_inputs) ⇒ Promise.<Object>
      • ._call(model_inputs) ⇒ Promise.<SamImageSegmentationOutput>
    • .SamImageSegmentationOutput
      • new SamImageSegmentationOutput(output)
    • .Wav2Vec2Model
    • .Wav2Vec2ForAudioFrameClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .PyAnnoteModel
    • .PyAnnoteForAudioFrameClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .UniSpeechModel
    • .UniSpeechForCTC
      • ._call(model_inputs)
    • .UniSpeechForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .UniSpeechSatModel
    • .UniSpeechSatForCTC
      • ._call(model_inputs)
    • .UniSpeechSatForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .UniSpeechSatForAudioFrameClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .Wav2Vec2BertModel
    • .Wav2Vec2BertForCTC
      • ._call(model_inputs)
    • .Wav2Vec2BertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .HubertModel
    • .HubertForCTC
      • ._call(model_inputs)
    • .HubertForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .WavLMPreTrainedModel
    • .WavLMModel
    • .WavLMForCTC
      • ._call(model_inputs)
    • .WavLMForSequenceClassification
      • ._call(model_inputs) ⇒ Promise.<SequenceClassifierOutput>
    • .WavLMForXVector
      • ._call(model_inputs) ⇒ Promise.<XVectorOutput>
    • .WavLMForAudioFrameClassification
      • ._call(model_inputs) ⇒ Promise.<TokenClassifierOutput>
    • .SpeechT5PreTrainedModel
    • .SpeechT5Model
    • .SpeechT5ForSpeechToText
    • .SpeechT5ForTextToSpeech
      • .generate_speech(input_values, speaker_embeddings, options) ⇒ Promise.<SpeechOutput>
    • .SpeechT5HifiGan
    • .TrOCRForCausalLM
    • .MistralPreTrainedModel
    • .Starcoder2PreTrainedModel
    • .FalconPreTrainedModel
    • .ClapTextModelWithProjection
      • .from_pretrained() : *
    • .ClapAudioModelWithProjection
      • .from_pretrained() : *
    • .VitsModel
      • ._call(model_inputs) ⇒ Promise.<VitsModelOutput>
    • .SegformerModel
    • .SegformerForImageClassification
    • .SegformerForSemanticSegmentation
    • .StableLmModel
    • .StableLmForCausalLM
    • .EfficientNetModel
    • .EfficientNetForImageClassification
      • ._call(model_inputs)
    • .MusicgenModel
    • .MusicgenForCausalLM
    • .MusicgenForConditionalGeneration
      • ._apply_and_filter_by_delay_pattern_mask(outputs) ⇒ Tensor
      • .generate(options) ⇒ Promise.<(ModelOutput|Tensor)>
    • .MobileNetV1Model
    • .MobileNetV1ForImageClassification
      • ._call(model_inputs)
    • .MobileNetV2Model
    • .MobileNetV2ForImageClassification
      • ._call(model_inputs)
    • .MobileNetV3Model
    • .MobileNetV3ForImageClassification
      • ._call(model_inputs)
    • .MobileNetV4Model
    • .MobileNetV4ForImageClassification
      • ._call(model_inputs)
    • .DecisionTransformerModel
    • .MgpstrForSceneTextRecognition
      • ._call(model_inputs)
    • .PatchTSTModel
    • .PatchTSTForPrediction
    • .PatchTSMixerModel
    • .PatchTSMixerForPrediction
    • .PretrainedMixin
      • instance
        • .MODEL_CLASS_MAPPINGS : *
        • .BASE_IF_FAIL
      • static
        • .from_pretrained() : *
    • .AutoModel
      • new AutoModel()
      • .MODEL_CLASS_MAPPINGS : *
    • .AutoModelForSequenceClassification
      • new AutoModelForSequenceClassification()
    • .AutoModelForTokenClassification
      • new AutoModelForTokenClassification()
    • .AutoModelForSeq2SeqLM
      • new AutoModelForSeq2SeqLM()
    • .AutoModelForSpeechSeq2Seq
      • new AutoModelForSpeechSeq2Seq()
    • .AutoModelForTextToSpectrogram
      • new AutoModelForTextToSpectrogram()
    • .AutoModelForTextToWaveform
      • new AutoModelForTextToWaveform()
    • .AutoModelForCausalLM
      • new AutoModelForCausalLM()
    • .AutoModelForMaskedLM
      • new AutoModelForMaskedLM()
    • .AutoModelForQuestionAnswering
      • new AutoModelForQuestionAnswering()
    • .AutoModelForVision2Seq
      • new AutoModelForVision2Seq()
    • .AutoModelForImageClassification
      • new AutoModelForImageClassification()
    • .AutoModelForImageSegmentation
      • new AutoModelForImageSegmentation()
    • .AutoModelForSemanticSegmentation
      • new AutoModelForSemanticSegmentation()
    • .AutoModelForUniversalSegmentation
      • new AutoModelForUniversalSegmentation()
    • .AutoModelForObjectDetection
      • new AutoModelForObjectDetection()
    • .AutoModelForMaskGeneration
      • new AutoModelForMaskGeneration()
    • .Seq2SeqLMOutput
      • new Seq2SeqLMOutput(output)
    • .SequenceClassifierOutput
      • new SequenceClassifierOutput(output)
    • .XVectorOutput
      • new XVectorOutput(output)
    • .TokenClassifierOutput
      • new TokenClassifierOutput(output)
    • .MaskedLMOutput
      • new MaskedLMOutput(output)
    • .QuestionAnsweringModelOutput
      • new QuestionAnsweringModelOutput(output)
    • .CausalLMOutput
      • new CausalLMOutput(output)
    • .CausalLMOutputWithPast
      • new CausalLMOutputWithPast(output)
    • .ImageMattingOutput
      • new ImageMattingOutput(output)
    • .VitsModelOutput
      • new VitsModelOutput(output)
  • inner
    • ~cumsum_masked_fill(attention_mask) ⇒ Object
    • ~createPositionIds()
    • ~SamModelInputs : Object
    • ~SpeechOutput : Object

models.PreTrainedModel

A base class for pre-trained models that provides the model configuration and an ONNX session.

Kind: static class of models

• .PreTrainedModel
  • new PreTrainedModel(config, sessions, configs)
  • instance
    • .custom_config : *
    • .generation_config ⇒ GenerationConfig | null
    • .dispose() ⇒ Promise.<Array<unknown>>
    • ._call(model_inputs) ⇒ Promise.<Object>
    • .forward(model_inputs) ⇒ Promise.<Object>
    • ._get_logits_warper(generation_config) ⇒ LogitsProcessorList
    • ._prepare_generation_config(generation_config, kwargs) ⇒ GenerationConfig
    • ._get_stopping_criteria(generation_config, [stopping_criteria])
    • ._validate_model_class()
    • ._update_model_kwargs_for_generation(inputs) ⇒ Object
    • ._prepare_model_inputs(params) ⇒ Object
    • ._prepare_decoder_input_ids_for_generation(param0)
    • .generate(options) ⇒ Promise.<(ModelOutput|Tensor)>
    • .getPastKeyValues(decoderResults, pastKeyValues) ⇒ Object
    • .getAttentions(model_output) ⇒ *
    • .addPastKeyValues(decoderFeeds, pastKeyValues)
  • static
    • .from_pretrained(pretrained_model_name_or_path, options) ⇒ Promise.<PreTrainedModel>

new PreTrainedModel(config, sessions, configs)

Creates a new instance of the PreTrainedModel class.

preTrainedModel.custom_config : <code> * </code>

Kind: instance property of PreTrainedModel

preTrainedModel.generation_config ⇒ <code> GenerationConfig </code> | <code> null </code>

Get the model’s generation config, if it exists.

Kind: instance property of PreTrainedModel
Returns: GenerationConfig | null - The model’s generation config if it exists, otherwise null.

preTrainedModel.dispose() ⇒ <code> Promise. < Array < unknown > > </code>

Disposes of all the ONNX sessions that were created during inference.

Kind: instance method of PreTrainedModel
Returns: Promise.<Array<unknown>> - An array of promises, one for each ONNX session that is being disposed.
Todo

• Use https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/FinalizationRegistry

preTrainedModel._call(model_inputs) ⇒ <code> Promise. < Object > </code>

Runs the model with the provided inputs

Kind: instance method of PreTrainedModel
Returns: Promise.<Object> - Object containing output tensors

preTrainedModel.forward(model_inputs) ⇒ <code> Promise. < Object > </code>

Forward method for a pretrained model. If not overridden by a subclass, the correct forward method will be chosen based on the model type.

Kind: instance method of PreTrainedModel
Returns: Promise.<Object> - The output data from the model in the format specified in the ONNX model.
Throws:

• Error This method must be implemented in subclasses.

preTrainedModel._get_logits_warper(generation_config) ⇒ <code> LogitsProcessorList </code>

This function returns a [LogitsProcessorList] list object that contains all relevant [LogitsWarper] instances used for multinomial sampling.

Kind: instance method of PreTrainedModel
Returns: LogitsProcessorList - generation_config

preTrainedModel._prepare_generation_config(generation_config, kwargs) ⇒ <code> GenerationConfig </code>

This function merges multiple generation configs together to form a final generation config to be used by the model for text generation. It first creates an empty GenerationConfig object, then it applies the model’s own generation_config property to it. Finally, if a generation_config object was passed in the arguments, it overwrites the corresponding properties in the final config with those of the passed config object.

Kind: instance method of PreTrainedModel
Returns: GenerationConfig - The final generation config object to be used by the model for text generation.

preTrainedModel._get_stopping_criteria(generation_config, [stopping_criteria])

Kind: instance method of PreTrainedModel

preTrainedModel._validate_model_class()

Confirms that the model class is compatible with generation. If not, raises an exception that points to the right class to use.

Kind: instance method of PreTrainedModel

preTrainedModel._update_model_kwargs_for_generation(inputs) ⇒ <code> Object </code>

Kind: instance method of PreTrainedModel
Returns: Object - The updated model inputs for the next generation iteration.

preTrainedModel._prepare_model_inputs(params) ⇒ <code> Object </code>

This function extracts the model-specific inputs for generation.

Kind: instance method of PreTrainedModel
Returns: Object - The model-specific inputs for generation.

preTrainedModel._prepare_decoder_input_ids_for_generation(param0)

Prepares decoder_input_ids for generation with encoder-decoder models

Kind: instance method of PreTrainedModel

preTrainedModel.generate(options) ⇒ <code> Promise. < (ModelOutput|Tensor) > </code>

Generates sequences of token ids for models with a language modeling head.

Kind: instance method of PreTrainedModel
Returns: Promise.<(ModelOutput|Tensor)> - The output of the model, which can contain the generated token ids, attentions, and scores.

preTrainedModel.getPastKeyValues(decoderResults, pastKeyValues) ⇒ <code> Object </code>

Returns an object containing past key values from the given decoder results object.

Kind: instance method of PreTrainedModel
Returns: Object - An object containing past key values.

preTrainedModel.getAttentions(model_output) ⇒ <code> * </code>

Returns an object containing attentions from the given model output object.

Kind: instance method of PreTrainedModel
Returns: * - An object containing attentions.

preTrainedModel.addPastKeyValues(decoderFeeds, pastKeyValues)

Adds past key values to the decoder feeds object. If pastKeyValues is null, creates new tensors for past key values.

Kind: instance method of PreTrainedModel

PreTrainedModel.from_pretrained(pretrained_model_name_or_path, options) ⇒ <code> Promise. < PreTrainedModel > </code>

Instantiate one of the model classes of the library from a pretrained model.

The model class to instantiate is selected based on the model_type property of the config object (either passed as an argument or loaded from pretrained_model_name_or_path if possible)

Kind: static method of PreTrainedModel
Returns: Promise.<PreTrainedModel> - A new instance of the PreTrainedModel class.

models.BaseModelOutput

Base class for model’s outputs, with potential hidden states and attentions.

Kind: static class of models

new BaseModelOutput(output)

models.BertForMaskedLM

BertForMaskedLM is a class representing a BERT model for masked language modeling.

Kind: static class of models

bertForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of BertForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.BertForSequenceClassification

BertForSequenceClassification is a class representing a BERT model for sequence classification.

Kind: static class of models

bertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of BertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.BertForTokenClassification

BertForTokenClassification is a class representing a BERT model for token classification.

Kind: static class of models

bertForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of BertForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.BertForQuestionAnswering

BertForQuestionAnswering is a class representing a BERT model for question answering.

Kind: static class of models

bertForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of BertForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.RoFormerModel

The bare RoFormer Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.RoFormerForMaskedLM

RoFormer Model with a language modeling head on top.

Kind: static class of models

roFormerForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of RoFormerForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.RoFormerForSequenceClassification

RoFormer Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

roFormerForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of RoFormerForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.RoFormerForTokenClassification

RoFormer Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

roFormerForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of RoFormerForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.RoFormerForQuestionAnswering

RoFormer Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

Kind: static class of models

roFormerForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of RoFormerForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.ConvBertModel

The bare ConvBERT Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.ConvBertForMaskedLM

ConvBERT Model with a language modeling head on top.

Kind: static class of models

convBertForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of ConvBertForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.ConvBertForSequenceClassification

ConvBERT Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

convBertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of ConvBertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.ConvBertForTokenClassification

ConvBERT Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

convBertForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of ConvBertForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.ConvBertForQuestionAnswering

ConvBERT Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits)

Kind: static class of models

convBertForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of ConvBertForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.ElectraModel

The bare Electra Model transformer outputting raw hidden-states without any specific head on top. Identical to the BERT model except that it uses an additional linear layer between the embedding layer and the encoder if the hidden size and embedding size are different.

Kind: static class of models

models.ElectraForMaskedLM

Electra model with a language modeling head on top.

Kind: static class of models

electraForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of ElectraForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.ElectraForSequenceClassification

ELECTRA Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

electraForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of ElectraForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.ElectraForTokenClassification

Electra model with a token classification head on top.

Kind: static class of models

electraForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of ElectraForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.ElectraForQuestionAnswering

LECTRA Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

Kind: static class of models

electraForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of ElectraForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.CamembertModel

The bare CamemBERT Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.CamembertForMaskedLM

CamemBERT Model with a language modeling head on top.

Kind: static class of models

camembertForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of CamembertForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.CamembertForSequenceClassification

CamemBERT Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output) e.g. for GLUE tasks.

Kind: static class of models

camembertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of CamembertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.CamembertForTokenClassification

CamemBERT Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

camembertForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of CamembertForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.CamembertForQuestionAnswering

CamemBERT Model with a span classification head on top for extractive question-answering tasks

Kind: static class of models

camembertForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of CamembertForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.DebertaModel

The bare DeBERTa Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.DebertaForMaskedLM

DeBERTa Model with a language modeling head on top.

Kind: static class of models

debertaForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.DebertaForSequenceClassification

DeBERTa Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

debertaForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.DebertaForTokenClassification

DeBERTa Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

debertaForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.DebertaForQuestionAnswering

DeBERTa Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

Kind: static class of models

debertaForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.DebertaV2Model

The bare DeBERTa-V2 Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.DebertaV2ForMaskedLM

DeBERTa-V2 Model with a language modeling head on top.

Kind: static class of models

debertaV2ForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaV2ForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.DebertaV2ForSequenceClassification

DeBERTa-V2 Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

debertaV2ForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaV2ForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.DebertaV2ForTokenClassification

DeBERTa-V2 Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

debertaV2ForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaV2ForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.DebertaV2ForQuestionAnswering

DeBERTa-V2 Model with a span classification head on top for extractive question-answering tasks like SQuAD (a linear layers on top of the hidden-states output to compute span start logits and span end logits).

Kind: static class of models

debertaV2ForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of DebertaV2ForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.DistilBertForSequenceClassification

DistilBertForSequenceClassification is a class representing a DistilBERT model for sequence classification.

Kind: static class of models

distilBertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DistilBertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.DistilBertForTokenClassification

DistilBertForTokenClassification is a class representing a DistilBERT model for token classification.

Kind: static class of models

distilBertForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of DistilBertForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.DistilBertForQuestionAnswering

DistilBertForQuestionAnswering is a class representing a DistilBERT model for question answering.

Kind: static class of models

distilBertForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of DistilBertForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.DistilBertForMaskedLM

DistilBertForMaskedLM is a class representing a DistilBERT model for masking task.

Kind: static class of models

distilBertForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of DistilBertForMaskedLM
Returns: Promise.<MaskedLMOutput> - returned object

models.EsmModel

The bare ESM Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.EsmForMaskedLM

ESM Model with a language modeling head on top.

Kind: static class of models

esmForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of EsmForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.EsmForSequenceClassification

ESM Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

esmForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of EsmForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.EsmForTokenClassification

ESM Model with a token classification head on top (a linear layer on top of the hidden-states output) e.g. for Named-Entity-Recognition (NER) tasks.

Kind: static class of models

esmForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of EsmForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.MobileBertForMaskedLM

MobileBertForMaskedLM is a class representing a MobileBERT model for masking task.

Kind: static class of models

mobileBertForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of MobileBertForMaskedLM
Returns: Promise.<MaskedLMOutput> - returned object

models.MobileBertForSequenceClassification

MobileBert Model transformer with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

mobileBertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of MobileBertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - returned object

models.MobileBertForQuestionAnswering

MobileBert Model with a span classification head on top for extractive question-answering tasks

Kind: static class of models

mobileBertForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of MobileBertForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - returned object

models.MPNetModel

The bare MPNet Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.MPNetForMaskedLM

MPNetForMaskedLM is a class representing a MPNet model for masked language modeling.

Kind: static class of models

mpNetForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of MPNetForMaskedLM
Returns: Promise.<MaskedLMOutput> - An object containing the model’s output logits for masked language modeling.

models.MPNetForSequenceClassification

MPNetForSequenceClassification is a class representing a MPNet model for sequence classification.

Kind: static class of models

mpNetForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of MPNetForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.MPNetForTokenClassification

MPNetForTokenClassification is a class representing a MPNet model for token classification.

Kind: static class of models

mpNetForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of MPNetForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.MPNetForQuestionAnswering

MPNetForQuestionAnswering is a class representing a MPNet model for question answering.

Kind: static class of models

mpNetForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of MPNetForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - An object containing the model’s output logits for question answering.

models.T5ForConditionalGeneration

T5Model is a class representing a T5 model for conditional generation.

Kind: static class of models

models.LongT5PreTrainedModel

An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.

Kind: static class of models

models.LongT5Model

The bare LONGT5 Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.LongT5ForConditionalGeneration

LONGT5 Model with a language modeling head on top.

Kind: static class of models

models.MT5ForConditionalGeneration

A class representing a conditional sequence-to-sequence model based on the MT5 architecture.

Kind: static class of models

models.BartModel

The bare BART Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.BartForConditionalGeneration

The BART Model with a language modeling head. Can be used for summarization.

Kind: static class of models

models.BartForSequenceClassification

Bart model with a sequence classification/head on top (a linear layer on top of the pooled output)

Kind: static class of models

bartForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of BartForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.MBartModel

The bare MBART Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.MBartForConditionalGeneration

The MBART Model with a language modeling head. Can be used for summarization, after fine-tuning the pretrained models.

Kind: static class of models

models.MBartForSequenceClassification

MBart model with a sequence classification/head on top (a linear layer on top of the pooled output).

Kind: static class of models

mBartForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of MBartForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.BlenderbotModel

The bare Blenderbot Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.BlenderbotForConditionalGeneration

The Blenderbot Model with a language modeling head. Can be used for summarization.

Kind: static class of models

models.BlenderbotSmallModel

The bare BlenderbotSmall Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.BlenderbotSmallForConditionalGeneration

The BlenderbotSmall Model with a language modeling head. Can be used for summarization.

Kind: static class of models

models.RobertaForMaskedLM

RobertaForMaskedLM class for performing masked language modeling on Roberta models.

Kind: static class of models

robertaForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of RobertaForMaskedLM
Returns: Promise.<MaskedLMOutput> - returned object

models.RobertaForSequenceClassification

RobertaForSequenceClassification class for performing sequence classification on Roberta models.

Kind: static class of models

robertaForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of RobertaForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - returned object

models.RobertaForTokenClassification

RobertaForTokenClassification class for performing token classification on Roberta models.

Kind: static class of models

robertaForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of RobertaForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.RobertaForQuestionAnswering

RobertaForQuestionAnswering class for performing question answering on Roberta models.

Kind: static class of models

robertaForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of RobertaForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - returned object

models.XLMPreTrainedModel

An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.

Kind: static class of models

models.XLMModel

The bare XLM Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.XLMWithLMHeadModel

The XLM Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

xlmWithLMHeadModel._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMWithLMHeadModel
Returns: Promise.<MaskedLMOutput> - returned object

models.XLMForSequenceClassification

XLM Model with a sequence classification/regression head on top (a linear layer on top of the pooled output)

Kind: static class of models

xlmForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - returned object

models.XLMForTokenClassification

XLM Model with a token classification head on top (a linear layer on top of the hidden-states output)

Kind: static class of models

xlmForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.XLMForQuestionAnswering

XLM Model with a span classification head on top for extractive question-answering tasks

Kind: static class of models

xlmForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - returned object

models.XLMRobertaForMaskedLM

XLMRobertaForMaskedLM class for performing masked language modeling on XLMRoberta models.

Kind: static class of models

xlmRobertaForMaskedLM._call(model_inputs) ⇒ <code> Promise. < MaskedLMOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMRobertaForMaskedLM
Returns: Promise.<MaskedLMOutput> - returned object

models.XLMRobertaForSequenceClassification

XLMRobertaForSequenceClassification class for performing sequence classification on XLMRoberta models.

Kind: static class of models

xlmRobertaForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMRobertaForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - returned object

models.XLMRobertaForTokenClassification

XLMRobertaForTokenClassification class for performing token classification on XLMRoberta models.

Kind: static class of models

xlmRobertaForTokenClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMRobertaForTokenClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for token classification.

models.XLMRobertaForQuestionAnswering

XLMRobertaForQuestionAnswering class for performing question answering on XLMRoberta models.

Kind: static class of models

xlmRobertaForQuestionAnswering._call(model_inputs) ⇒ <code> Promise. < QuestionAnsweringModelOutput > </code>

Calls the model on new inputs.

Kind: instance method of XLMRobertaForQuestionAnswering
Returns: Promise.<QuestionAnsweringModelOutput> - returned object

models.ASTModel

The bare AST Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.ASTForAudioClassification

Audio Spectrogram Transformer model with an audio classification head on top (a linear layer on top of the pooled output) e.g. for datasets like AudioSet, Speech Commands v2.

Kind: static class of models

models.WhisperModel

WhisperModel class for training Whisper models without a language model head.

Kind: static class of models

models.WhisperForConditionalGeneration

WhisperForConditionalGeneration class for generating conditional outputs from Whisper models.

Kind: static class of models

• .WhisperForConditionalGeneration
  • ._retrieve_init_tokens(generation_config)
  • .generate(options) ⇒ Promise.<(ModelOutput|Tensor)>
  • ._extract_token_timestamps(generate_outputs, alignment_heads, [num_frames], [time_precision]) ⇒ Tensor

whisperForConditionalGeneration._retrieve_init_tokens(generation_config)

Kind: instance method of WhisperForConditionalGeneration

whisperForConditionalGeneration.generate(options) ⇒ <code> Promise. < (ModelOutput|Tensor) > </code>

Transcribes or translates log-mel input features to a sequence of auto-regressively generated token ids.

Kind: instance method of WhisperForConditionalGeneration
Returns: Promise.<(ModelOutput|Tensor)> - The output of the model, which can contain the generated token ids, attentions, and scores.

whisperForConditionalGeneration._extract_token_timestamps(generate_outputs, alignment_heads, [num_frames], [time_precision]) ⇒ <code> Tensor </code>

Calculates token-level timestamps using the encoder-decoder cross-attentions and dynamic time-warping (DTW) to map each output token to a position in the input audio. If num_frames is specified, the encoder-decoder cross-attentions will be cropped before applying DTW.

Kind: instance method of WhisperForConditionalGeneration
Returns: Tensor - tensor containing the timestamps in seconds for each predicted token

models.MoonshineModel

MoonshineModel class for training Moonshine models without a language model head.

Kind: static class of models

models.VisionEncoderDecoderModel

Vision Encoder-Decoder model based on OpenAI’s GPT architecture for image captioning and other vision tasks

Kind: static class of models

models.LlavaForConditionalGeneration

The LLAVA model which consists of a vision backbone and a language model.

Kind: static class of models

models.Idefics3ForConditionalGeneration

The LLAVA model which consists of a vision backbone and a language model.

Kind: static class of models

models.CLIPModel

CLIP Text and Vision Model with a projection layers on top

Example: Perform zero-shot image classification with a CLIPModel.

import { AutoTokenizer, AutoProcessor, CLIPModel, RawImage } from '@huggingface/transformers';

// Load tokenizer, processor, and model
let tokenizer = await AutoTokenizer.from_pretrained('Xenova/clip-vit-base-patch16');
let processor = await AutoProcessor.from_pretrained('Xenova/clip-vit-base-patch16');
let model = await CLIPModel.from_pretrained('Xenova/clip-vit-base-patch16');

// Run tokenization
let texts = ['a photo of a car', 'a photo of a football match']
let text_inputs = tokenizer(texts, { padding: true, truncation: true });

// Read image and run processor
let image = await RawImage.read('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/football-match.jpg');
let image_inputs = await processor(image);

// Run model with both text and pixel inputs
let output = await model({ ...text_inputs, ...image_inputs });
// {
//   logits_per_image: Tensor {
//     dims: [ 1, 2 ],
//     data: Float32Array(2) [ 18.579734802246094, 24.31830596923828 ],
//   },
//   logits_per_text: Tensor {
//     dims: [ 2, 1 ],
//     data: Float32Array(2) [ 18.579734802246094, 24.31830596923828 ],
//   },
//   text_embeds: Tensor {
//     dims: [ 2, 512 ],
//     data: Float32Array(1024) [ ... ],
//   },
//   image_embeds: Tensor {
//     dims: [ 1, 512 ],
//     data: Float32Array(512) [ ... ],
//   }
// }

Kind: static class of models

models.CLIPTextModel

The text model from CLIP without any head or projection on top.

Kind: static class of models

CLIPTextModel.from_pretrained() : <code> * </code>

Kind: static method of CLIPTextModel

models.CLIPTextModelWithProjection

CLIP Text Model with a projection layer on top (a linear layer on top of the pooled output)

Example: Compute text embeddings with CLIPTextModelWithProjection.

import { AutoTokenizer, CLIPTextModelWithProjection } from '@huggingface/transformers';

// Load tokenizer and text model
const tokenizer = await AutoTokenizer.from_pretrained('Xenova/clip-vit-base-patch16');
const text_model = await CLIPTextModelWithProjection.from_pretrained('Xenova/clip-vit-base-patch16');

// Run tokenization
let texts = ['a photo of a car', 'a photo of a football match'];
let text_inputs = tokenizer(texts, { padding: true, truncation: true });

// Compute embeddings
const { text_embeds } = await text_model(text_inputs);
// Tensor {
//   dims: [ 2, 512 ],
//   type: 'float32',
//   data: Float32Array(1024) [ ... ],
//   size: 1024
// }

Kind: static class of models

CLIPTextModelWithProjection.from_pretrained() : <code> * </code>

Kind: static method of CLIPTextModelWithProjection

models.CLIPVisionModel

The vision model from CLIP without any head or projection on top.

Kind: static class of models

CLIPVisionModel.from_pretrained() : <code> * </code>

Kind: static method of CLIPVisionModel

models.CLIPVisionModelWithProjection

CLIP Vision Model with a projection layer on top (a linear layer on top of the pooled output)

Example: Compute vision embeddings with CLIPVisionModelWithProjection.

import { AutoProcessor, CLIPVisionModelWithProjection, RawImage} from '@huggingface/transformers';

// Load processor and vision model
const processor = await AutoProcessor.from_pretrained('Xenova/clip-vit-base-patch16');
const vision_model = await CLIPVisionModelWithProjection.from_pretrained('Xenova/clip-vit-base-patch16');

// Read image and run processor
let image = await RawImage.read('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/football-match.jpg');
let image_inputs = await processor(image);

// Compute embeddings
const { image_embeds } = await vision_model(image_inputs);
// Tensor {
//   dims: [ 1, 512 ],
//   type: 'float32',
//   data: Float32Array(512) [ ... ],
//   size: 512
// }

Kind: static class of models

CLIPVisionModelWithProjection.from_pretrained() : <code> * </code>

Kind: static method of CLIPVisionModelWithProjection

models.SiglipModel

SigLIP Text and Vision Model with a projection layers on top

Example: Perform zero-shot image classification with a SiglipModel.

import { AutoTokenizer, AutoProcessor, SiglipModel, RawImage } from '@huggingface/transformers';

// Load tokenizer, processor, and model
const tokenizer = await AutoTokenizer.from_pretrained('Xenova/siglip-base-patch16-224');
const processor = await AutoProcessor.from_pretrained('Xenova/siglip-base-patch16-224');
const model = await SiglipModel.from_pretrained('Xenova/siglip-base-patch16-224');

// Run tokenization
const texts = ['a photo of 2 cats', 'a photo of 2 dogs'];
const text_inputs = tokenizer(texts, { padding: 'max_length', truncation: true });

// Read image and run processor
const image = await RawImage.read('http://images.cocodataset.org/val2017/000000039769.jpg');
const image_inputs = await processor(image);

// Run model with both text and pixel inputs
const output = await model({ ...text_inputs, ...image_inputs });
// {
//   logits_per_image: Tensor {
//     dims: [ 1, 2 ],
//     data: Float32Array(2) [ -1.6019744873046875, -10.720091819763184 ],
//   },
//   logits_per_text: Tensor {
//     dims: [ 2, 1 ],
//     data: Float32Array(2) [ -1.6019744873046875, -10.720091819763184 ],
//   },
//   text_embeds: Tensor {
//     dims: [ 2, 768 ],
//     data: Float32Array(1536) [ ... ],
//   },
//   image_embeds: Tensor {
//     dims: [ 1, 768 ],
//     data: Float32Array(768) [ ... ],
//   }
// }

Kind: static class of models

models.SiglipTextModel

The text model from SigLIP without any head or projection on top.

Example: Compute text embeddings with SiglipTextModel.

import { AutoTokenizer, SiglipTextModel } from '@huggingface/transformers';

// Load tokenizer and text model
const tokenizer = await AutoTokenizer.from_pretrained('Xenova/siglip-base-patch16-224');
const text_model = await SiglipTextModel.from_pretrained('Xenova/siglip-base-patch16-224');

// Run tokenization
const texts = ['a photo of 2 cats', 'a photo of 2 dogs'];
const text_inputs = tokenizer(texts, { padding: 'max_length', truncation: true });

// Compute embeddings
const { pooler_output } = await text_model(text_inputs);
// Tensor {
//   dims: [ 2, 768 ],
//   type: 'float32',
//   data: Float32Array(1536) [ ... ],
//   size: 1536
// }

Kind: static class of models

SiglipTextModel.from_pretrained() : <code> * </code>

Kind: static method of SiglipTextModel

models.SiglipVisionModel

The vision model from SigLIP without any head or projection on top.

Example: Compute vision embeddings with SiglipVisionModel.

import { AutoProcessor, SiglipVisionModel, RawImage} from '@huggingface/transformers';

// Load processor and vision model
const processor = await AutoProcessor.from_pretrained('Xenova/siglip-base-patch16-224');
const vision_model = await SiglipVisionModel.from_pretrained('Xenova/siglip-base-patch16-224');

// Read image and run processor
const image = await RawImage.read('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/football-match.jpg');
const image_inputs = await processor(image);

// Compute embeddings
const { pooler_output } = await vision_model(image_inputs);
// Tensor {
//   dims: [ 1, 768 ],
//   type: 'float32',
//   data: Float32Array(768) [ ... ],
//   size: 768
// }

Kind: static class of models

SiglipVisionModel.from_pretrained() : <code> * </code>

Kind: static method of SiglipVisionModel

models.CLIPSegForImageSegmentation

CLIPSeg model with a Transformer-based decoder on top for zero-shot and one-shot image segmentation.

Example: Perform zero-shot image segmentation with a CLIPSegForImageSegmentation model.

import { AutoTokenizer, AutoProcessor, CLIPSegForImageSegmentation, RawImage } from '@huggingface/transformers';

// Load tokenizer, processor, and model
const tokenizer = await AutoTokenizer.from_pretrained('Xenova/clipseg-rd64-refined');
const processor = await AutoProcessor.from_pretrained('Xenova/clipseg-rd64-refined');
const model = await CLIPSegForImageSegmentation.from_pretrained('Xenova/clipseg-rd64-refined');

// Run tokenization
const texts = ['a glass', 'something to fill', 'wood', 'a jar'];
const text_inputs = tokenizer(texts, { padding: true, truncation: true });

// Read image and run processor
const image = await RawImage.read('https://github.com/timojl/clipseg/blob/master/example_image.jpg?raw=true');
const image_inputs = await processor(image);

// Run model with both text and pixel inputs
const { logits } = await model({ ...text_inputs, ...image_inputs });
// logits: Tensor {
//   dims: [4, 352, 352],
//   type: 'float32',
//   data: Float32Array(495616) [ ... ],
//   size: 495616
// }

You can visualize the predictions as follows:

const preds = logits
  .unsqueeze_(1)
  .sigmoid_()
  .mul_(255)
  .round_()
  .to('uint8');

for (let i = 0; i < preds.dims[0]; ++i) {
  const img = RawImage.fromTensor(preds[i]);
  img.save(`prediction_${i}.png`);
}

Kind: static class of models

models.GPT2LMHeadModel

GPT-2 language model head on top of the GPT-2 base model. This model is suitable for text generation tasks.

Kind: static class of models

models.JAISModel

The bare JAIS Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.JAISLMHeadModel

The JAIS Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

models.CodeGenModel

CodeGenModel is a class representing a code generation model without a language model head.

Kind: static class of models

models.CodeGenForCausalLM

CodeGenForCausalLM is a class that represents a code generation model based on the GPT-2 architecture. It extends the CodeGenPreTrainedModel class.

Kind: static class of models

models.LlamaPreTrainedModel

The bare LLama Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.LlamaModel

The bare LLaMA Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.CoherePreTrainedModel

The bare Cohere Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.GemmaPreTrainedModel

The bare Gemma Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.GemmaModel

The bare Gemma Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Gemma2PreTrainedModel

The bare Gemma2 Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Gemma2Model

The bare Gemma2 Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Qwen2PreTrainedModel

The bare Qwen2 Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Qwen2Model

The bare Qwen2 Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.PhiModel

The bare Phi Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Phi3Model

The bare Phi3 Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.BloomPreTrainedModel

The Bloom Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

models.BloomModel

The bare Bloom Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.BloomForCausalLM

The Bloom Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

models.MptModel

The bare Mpt Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.MptForCausalLM

The MPT Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

models.OPTModel

The bare OPT Model outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.OPTForCausalLM

The OPT Model transformer with a language modeling head on top (linear layer with weights tied to the input embeddings).

Kind: static class of models

models.VitPoseForPoseEstimation

The VitPose model with a pose estimation head on top.

Kind: static class of models

models.VitMatteForImageMatting

ViTMatte framework leveraging any vision backbone e.g. for ADE20k, CityScapes.

Example: Perform image matting with a VitMatteForImageMatting model.

import { AutoProcessor, VitMatteForImageMatting, RawImage } from '@huggingface/transformers';

// Load processor and model
const processor = await AutoProcessor.from_pretrained('Xenova/vitmatte-small-distinctions-646');
const model = await VitMatteForImageMatting.from_pretrained('Xenova/vitmatte-small-distinctions-646');

// Load image and trimap
const image = await RawImage.fromURL('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/vitmatte_image.png');
const trimap = await RawImage.fromURL('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/vitmatte_trimap.png');

// Prepare image + trimap for the model
const inputs = await processor(image, trimap);

// Predict alpha matte
const { alphas } = await model(inputs);
// Tensor {
//   dims: [ 1, 1, 640, 960 ],
//   type: 'float32',
//   size: 614400,
//   data: Float32Array(614400) [ 0.9894027709960938, 0.9970508813858032, ... ]
// }

You can visualize the alpha matte as follows:

import { Tensor, cat } from '@huggingface/transformers';

// Visualize predicted alpha matte
const imageTensor = image.toTensor();

// Convert float (0-1) alpha matte to uint8 (0-255)
const alphaChannel = alphas
  .squeeze(0)
  .mul_(255)
  .clamp_(0, 255)
  .round_()
  .to('uint8');

// Concatenate original image with predicted alpha
const imageData = cat([imageTensor, alphaChannel], 0);

// Save output image
const outputImage = RawImage.fromTensor(imageData);
outputImage.save('output.png');

Kind: static class of models

vitMatteForImageMatting._call(model_inputs)

Kind: instance method of VitMatteForImageMatting

models.DetrObjectDetectionOutput

Kind: static class of models

new DetrObjectDetectionOutput(output)

models.DetrSegmentationOutput

Kind: static class of models

new DetrSegmentationOutput(output)

models.RTDetrObjectDetectionOutput

Kind: static class of models

new RTDetrObjectDetectionOutput(output)

models.TableTransformerModel

The bare Table Transformer Model (consisting of a backbone and encoder-decoder Transformer) outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.TableTransformerForObjectDetection

Table Transformer Model (consisting of a backbone and encoder-decoder Transformer) with object detection heads on top, for tasks such as COCO detection.

Kind: static class of models

tableTransformerForObjectDetection._call(model_inputs)

Kind: instance method of TableTransformerForObjectDetection

models.ResNetPreTrainedModel

An abstract class to handle weights initialization and a simple interface for downloading and loading pretrained models.

Kind: static class of models

models.ResNetModel

The bare ResNet model outputting raw features without any specific head on top.

Kind: static class of models

models.ResNetForImageClassification

ResNet Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet.

Kind: static class of models

resNetForImageClassification._call(model_inputs)

Kind: instance method of ResNetForImageClassification

models.Swin2SRModel

The bare Swin2SR Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Swin2SRForImageSuperResolution

Swin2SR Model transformer with an upsampler head on top for image super resolution and restoration.

Example: Super-resolution w/ Xenova/swin2SR-classical-sr-x2-64.

import { AutoProcessor, Swin2SRForImageSuperResolution, RawImage } from '@huggingface/transformers';

// Load processor and model
const model_id = 'Xenova/swin2SR-classical-sr-x2-64';
const processor = await AutoProcessor.from_pretrained(model_id);
const model = await Swin2SRForImageSuperResolution.from_pretrained(model_id);

// Prepare model inputs
const url = 'https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/butterfly.jpg';
const image = await RawImage.fromURL(url);
const inputs = await processor(image);

// Run model
const outputs = await model(inputs);

// Convert Tensor to RawImage
const output = outputs.reconstruction.squeeze().clamp_(0, 1).mul_(255).round_().to('uint8');
const outputImage = RawImage.fromTensor(output);
// RawImage {
//   data: Uint8Array(786432) [ 41, 31, 24, ... ],
//   width: 512,
//   height: 512,
//   channels: 3
// }

Kind: static class of models

models.DPTModel

The bare DPT Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.DPTForDepthEstimation

DPT Model with a depth estimation head on top (consisting of 3 convolutional layers) e.g. for KITTI, NYUv2.

Example: Depth estimation w/ Xenova/dpt-hybrid-midas.

import { DPTForDepthEstimation, AutoProcessor, RawImage, interpolate, max } from '@huggingface/transformers';

// Load model and processor
const model_id = 'Xenova/dpt-hybrid-midas';
const model = await DPTForDepthEstimation.from_pretrained(model_id);
const processor = await AutoProcessor.from_pretrained(model_id);

// Load image from URL
const url = 'http://images.cocodataset.org/val2017/000000039769.jpg';
const image = await RawImage.fromURL(url);

// Prepare image for the model
const inputs = await processor(image);

// Run model
const { predicted_depth } = await model(inputs);

// Interpolate to original size
const prediction = interpolate(predicted_depth, image.size.reverse(), 'bilinear', false);

// Visualize the prediction
const formatted = prediction.mul_(255 / max(prediction.data)[0]).to('uint8');
const depth = RawImage.fromTensor(formatted);
// RawImage {
//   data: Uint8Array(307200) [ 85, 85, 84, ... ],
//   width: 640,
//   height: 480,
//   channels: 1
// }

Kind: static class of models

models.DepthAnythingForDepthEstimation

Depth Anything Model with a depth estimation head on top (consisting of 3 convolutional layers) e.g. for KITTI, NYUv2.

Kind: static class of models

models.GLPNModel

The bare GLPN encoder (Mix-Transformer) outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.GLPNForDepthEstimation

GLPN Model transformer with a lightweight depth estimation head on top e.g. for KITTI, NYUv2.

Example: Depth estimation w/ Xenova/glpn-kitti.

import { GLPNForDepthEstimation, AutoProcessor, RawImage, interpolate, max } from '@huggingface/transformers';

// Load model and processor
const model_id = 'Xenova/glpn-kitti';
const model = await GLPNForDepthEstimation.from_pretrained(model_id);
const processor = await AutoProcessor.from_pretrained(model_id);

// Load image from URL
const url = 'http://images.cocodataset.org/val2017/000000039769.jpg';
const image = await RawImage.fromURL(url);

// Prepare image for the model
const inputs = await processor(image);

// Run model
const { predicted_depth } = await model(inputs);

// Interpolate to original size
const prediction = interpolate(predicted_depth, image.size.reverse(), 'bilinear', false);

// Visualize the prediction
const formatted = prediction.mul_(255 / max(prediction.data)[0]).to('uint8');
const depth = RawImage.fromTensor(formatted);
// RawImage {
//   data: Uint8Array(307200) [ 207, 169, 154, ... ],
//   width: 640,
//   height: 480,
//   channels: 1
// }

Kind: static class of models

models.DonutSwinModel

The bare Donut Swin Model transformer outputting raw hidden-states without any specific head on top.

Example: Step-by-step Document Parsing.

import { AutoProcessor, AutoTokenizer, AutoModelForVision2Seq, RawImage } from '@huggingface/transformers';

// Choose model to use
const model_id = 'Xenova/donut-base-finetuned-cord-v2';

// Prepare image inputs
const processor = await AutoProcessor.from_pretrained(model_id);
const url = 'https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/receipt.png';
const image = await RawImage.read(url);
const image_inputs = await processor(image);

// Prepare decoder inputs
const tokenizer = await AutoTokenizer.from_pretrained(model_id);
const task_prompt = '<s_cord-v2>';
const decoder_input_ids = tokenizer(task_prompt, {
  add_special_tokens: false,
}).input_ids;

// Create the model
const model = await AutoModelForVision2Seq.from_pretrained(model_id);

// Run inference
const output = await model.generate(image_inputs.pixel_values, {
  decoder_input_ids,
  max_length: model.config.decoder.max_position_embeddings,
});

// Decode output
const decoded = tokenizer.batch_decode(output)[0];
// <s_cord-v2><s_menu><s_nm> CINNAMON SUGAR</s_nm><s_unitprice> 17,000</s_unitprice><s_cnt> 1 x</s_cnt><s_price> 17,000</s_price></s_menu><s_sub_total><s_subtotal_price> 17,000</s_subtotal_price></s_sub_total><s_total><s_total_price> 17,000</s_total_price><s_cashprice> 20,000</s_cashprice><s_changeprice> 3,000</s_changeprice></s_total></s>

Example: Step-by-step Document Visual Question Answering (DocVQA)

import { AutoProcessor, AutoTokenizer, AutoModelForVision2Seq, RawImage } from '@huggingface/transformers';

// Choose model to use
const model_id = 'Xenova/donut-base-finetuned-docvqa';

// Prepare image inputs
const processor = await AutoProcessor.from_pretrained(model_id);
const url = 'https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/invoice.png';
const image = await RawImage.read(url);
const image_inputs = await processor(image);

// Prepare decoder inputs
const tokenizer = await AutoTokenizer.from_pretrained(model_id);
const question = 'What is the invoice number?';
const task_prompt = `<s_docvqa><s_question>${question}</s_question><s_answer>`;
const decoder_input_ids = tokenizer(task_prompt, {
  add_special_tokens: false,
}).input_ids;

// Create the model
const model = await AutoModelForVision2Seq.from_pretrained(model_id);

// Run inference
const output = await model.generate(image_inputs.pixel_values, {
  decoder_input_ids,
  max_length: model.config.decoder.max_position_embeddings,
});

// Decode output
const decoded = tokenizer.batch_decode(output)[0];
// <s_docvqa><s_question> What is the invoice number?</s_question><s_answer> us-001</s_answer></s>

Kind: static class of models

models.ConvNextModel

The bare ConvNext model outputting raw features without any specific head on top.

Kind: static class of models

models.ConvNextForImageClassification

ConvNext Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet.

Kind: static class of models

convNextForImageClassification._call(model_inputs)

Kind: instance method of ConvNextForImageClassification

models.ConvNextV2Model

The bare ConvNextV2 model outputting raw features without any specific head on top.

Kind: static class of models

models.ConvNextV2ForImageClassification

ConvNextV2 Model with an image classification head on top (a linear layer on top of the pooled features), e.g. for ImageNet.

Kind: static class of models

convNextV2ForImageClassification._call(model_inputs)

Kind: instance method of ConvNextV2ForImageClassification

models.Dinov2Model

The bare DINOv2 Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Dinov2ForImageClassification

Dinov2 Model transformer with an image classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for ImageNet.

Kind: static class of models

dinov2ForImageClassification._call(model_inputs)

Kind: instance method of Dinov2ForImageClassification

models.Dinov2WithRegistersModel

The bare Dinov2WithRegisters Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Dinov2WithRegistersForImageClassification

Dinov2WithRegisters Model transformer with an image classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for ImageNet.

Kind: static class of models

dinov2WithRegistersForImageClassification._call(model_inputs)

Kind: instance method of Dinov2WithRegistersForImageClassification

models.YolosObjectDetectionOutput

Kind: static class of models

new YolosObjectDetectionOutput(output)

models.SamModel

Segment Anything Model (SAM) for generating segmentation masks, given an input image and optional 2D location and bounding boxes.

Example: Perform mask generation w/ Xenova/sam-vit-base.

import { SamModel, AutoProcessor, RawImage } from '@huggingface/transformers';

const model = await SamModel.from_pretrained('Xenova/sam-vit-base');
const processor = await AutoProcessor.from_pretrained('Xenova/sam-vit-base');

const img_url = 'https://huggingface.co./ybelkada/segment-anything/resolve/main/assets/car.png';
const raw_image = await RawImage.read(img_url);
const input_points = [[[450, 600]]] // 2D localization of a window

const inputs = await processor(raw_image, { input_points });
const outputs = await model(inputs);

const masks = await processor.post_process_masks(outputs.pred_masks, inputs.original_sizes, inputs.reshaped_input_sizes);
// [
//   Tensor {
//     dims: [ 1, 3, 1764, 2646 ],
//     type: 'bool',
//     data: Uint8Array(14002632) [ ... ],
//     size: 14002632
//   }
// ]
const scores = outputs.iou_scores;
// Tensor {
//   dims: [ 1, 1, 3 ],
//   type: 'float32',
//   data: Float32Array(3) [
//     0.8892380595207214,
//     0.9311248064041138,
//     0.983696699142456
//   ],
//   size: 3
// }

Kind: static class of models

• .SamModel
  • .get_image_embeddings(model_inputs) ⇒ Promise.<{image_embeddings: Tensor, image_positional_embeddings: Tensor}>
  • .forward(model_inputs) ⇒ Promise.<Object>
  • ._call(model_inputs) ⇒ Promise.<SamImageSegmentationOutput>

samModel.get_image_embeddings(model_inputs) ⇒ <code> Promise. < {image_embeddings: Tensor, image_positional_embeddings: Tensor} > </code>

Compute image embeddings and positional image embeddings, given the pixel values of an image.

Kind: instance method of SamModel
Returns: Promise.<{image_embeddings: Tensor, image_positional_embeddings: Tensor}> - The image embeddings and positional image embeddings.

samModel.forward(model_inputs) ⇒ <code> Promise. < Object > </code>

Kind: instance method of SamModel
Returns: Promise.<Object> - The output of the model.

samModel._call(model_inputs) ⇒ <code> Promise. < SamImageSegmentationOutput > </code>

Runs the model with the provided inputs

Kind: instance method of SamModel
Returns: Promise.<SamImageSegmentationOutput> - Object containing segmentation outputs

models.SamImageSegmentationOutput

Base class for Segment-Anything model’s output.

Kind: static class of models

new SamImageSegmentationOutput(output)

models.Wav2Vec2Model

The bare Wav2Vec2 Model transformer outputting raw hidden-states without any specific head on top.

Example: Load and run a Wav2Vec2Model for feature extraction.

import { AutoProcessor, AutoModel, read_audio } from '@huggingface/transformers';

// Read and preprocess audio
const processor = await AutoProcessor.from_pretrained('Xenova/mms-300m');
const audio = await read_audio('https://huggingface.co./datasets/Narsil/asr_dummy/resolve/main/mlk.flac', 16000);
const inputs = await processor(audio);

// Run model with inputs
const model = await AutoModel.from_pretrained('Xenova/mms-300m');
const output = await model(inputs);
// {
//   last_hidden_state: Tensor {
//     dims: [ 1, 1144, 1024 ],
//     type: 'float32',
//     data: Float32Array(1171456) [ ... ],
//     size: 1171456
//   }
// }

Kind: static class of models

models.Wav2Vec2ForAudioFrameClassification

Wav2Vec2 Model with a frame classification head on top for tasks like Speaker Diarization.

Kind: static class of models

wav2Vec2ForAudioFrameClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of Wav2Vec2ForAudioFrameClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.PyAnnoteModel

The bare PyAnnote Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.PyAnnoteForAudioFrameClassification

PyAnnote Model with a frame classification head on top for tasks like Speaker Diarization.

Example: Load and run a PyAnnoteForAudioFrameClassification for speaker diarization.

import { AutoProcessor, AutoModelForAudioFrameClassification, read_audio } from '@huggingface/transformers';

// Load model and processor
const model_id = 'onnx-community/pyannote-segmentation-3.0';
const model = await AutoModelForAudioFrameClassification.from_pretrained(model_id);
const processor = await AutoProcessor.from_pretrained(model_id);

// Read and preprocess audio
const url = 'https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/mlk.wav';
const audio = await read_audio(url, processor.feature_extractor.config.sampling_rate);
const inputs = await processor(audio);

// Run model with inputs
const { logits } = await model(inputs);
// {
//   logits: Tensor {
//     dims: [ 1, 767, 7 ],  // [batch_size, num_frames, num_classes]
//     type: 'float32',
//     data: Float32Array(5369) [ ... ],
//     size: 5369
//   }
// }

const result = processor.post_process_speaker_diarization(logits, audio.length);
// [
//   [
//     { id: 0, start: 0, end: 1.0512535626298245, confidence: 0.8220156481664611 },
//     { id: 2, start: 1.0512535626298245, end: 2.3398869619825127, confidence: 0.9008811707860472 },
//     ...
//   ]
// ]

// Display result
console.table(result[0], ['start', 'end', 'id', 'confidence']);
// ┌─────────┬────────────────────┬────────────────────┬────┬─────────────────────┐
// │ (index) │ start              │ end                │ id │ confidence          │
// ├─────────┼────────────────────┼────────────────────┼────┼─────────────────────┤
// │ 0       │ 0                  │ 1.0512535626298245 │ 0  │ 0.8220156481664611  │
// │ 1       │ 1.0512535626298245 │ 2.3398869619825127 │ 2  │ 0.9008811707860472  │
// │ 2       │ 2.3398869619825127 │ 3.5946089560890773 │ 0  │ 0.7521651315796233  │
// │ 3       │ 3.5946089560890773 │ 4.578039708226655  │ 2  │ 0.8491978128022479  │
// │ 4       │ 4.578039708226655  │ 4.594995410849717  │ 0  │ 0.2935352600416393  │
// │ 5       │ 4.594995410849717  │ 6.121008646925269  │ 3  │ 0.6788051309866024  │
// │ 6       │ 6.121008646925269  │ 6.256654267909762  │ 0  │ 0.37125512393851134 │
// │ 7       │ 6.256654267909762  │ 8.630452635138397  │ 2  │ 0.7467035186353542  │
// │ 8       │ 8.630452635138397  │ 10.088643060721703 │ 0  │ 0.7689364814666032  │
// │ 9       │ 10.088643060721703 │ 12.58113134631177  │ 2  │ 0.9123324509131324  │
// │ 10      │ 12.58113134631177  │ 13.005023911888312 │ 0  │ 0.4828358177572041  │
// └─────────┴────────────────────┴────────────────────┴────┴─────────────────────┘

Kind: static class of models

pyAnnoteForAudioFrameClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of PyAnnoteForAudioFrameClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.UniSpeechModel

The bare UniSpeech Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.UniSpeechForCTC

UniSpeech Model with a language modeling head on top for Connectionist Temporal Classification (CTC).

Kind: static class of models

uniSpeechForCTC._call(model_inputs)

Kind: instance method of UniSpeechForCTC

models.UniSpeechForSequenceClassification

UniSpeech Model with a sequence classification head on top (a linear layer over the pooled output).

Kind: static class of models

uniSpeechForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of UniSpeechForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.UniSpeechSatModel

The bare UniSpeechSat Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.UniSpeechSatForCTC

UniSpeechSat Model with a language modeling head on top for Connectionist Temporal Classification (CTC).

Kind: static class of models

uniSpeechSatForCTC._call(model_inputs)

Kind: instance method of UniSpeechSatForCTC

models.UniSpeechSatForSequenceClassification

UniSpeechSat Model with a sequence classification head on top (a linear layer over the pooled output).

Kind: static class of models

uniSpeechSatForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of UniSpeechSatForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.UniSpeechSatForAudioFrameClassification

UniSpeechSat Model with a frame classification head on top for tasks like Speaker Diarization.

Kind: static class of models

uniSpeechSatForAudioFrameClassification._call(model_inputs) ⇒ <code> Promise. < TokenClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of UniSpeechSatForAudioFrameClassification
Returns: Promise.<TokenClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.Wav2Vec2BertModel

The bare Wav2Vec2Bert Model transformer outputting raw hidden-states without any specific head on top.

Kind: static class of models

models.Wav2Vec2BertForCTC

Wav2Vec2Bert Model with a language modeling head on top for Connectionist Temporal Classification (CTC).

Kind: static class of models

wav2Vec2BertForCTC._call(model_inputs)

Kind: instance method of Wav2Vec2BertForCTC

models.Wav2Vec2BertForSequenceClassification

Wav2Vec2Bert Model with a sequence classification head on top (a linear layer over the pooled output).

Kind: static class of models

wav2Vec2BertForSequenceClassification._call(model_inputs) ⇒ <code> Promise. < SequenceClassifierOutput > </code>

Calls the model on new inputs.

Kind: instance method of Wav2Vec2BertForSequenceClassification
Returns: Promise.<SequenceClassifierOutput> - An object containing the model’s output logits for sequence classification.

models.HubertModel

The bare Hubert Model transformer outputting raw hidden-states without any specific head on top.

Example: Load and run a HubertModel for feature extraction.

import { AutoProcessor, AutoModel, read_audio } from '@huggingface/transformers';

// Read and preprocess audio
const processor = await AutoProcessor.from_pretrained('Xenova/hubert-base-ls960');
const audio = await read_audio('https://huggingface.co./datasets/Xenova/transformers.js-docs/resolve/main/jfk.wav', 16000);
const inputs = await processor(audio);

// Load and run model with inputs
const model = await AutoModel.from_pretrained('Xenova/hubert-base-ls960');
const output = await model(inputs);
// {
//   last_hidden_state: Tensor {
//     dims: [ 1, 549, 768 ],
//     type: 'float32',
//     data: Float32Array(421632) [0.0682469978928566, 0.08104046434164047, -0.4975186586380005, ...],
//     size: 421632
//   }
// }

Kind: static class of models

models.HubertForCTC

Hubert Model with a language modeling head on top for Connectionist Temporal Classification (CTC).

Kind: static class of models

hubertForCTC._call(model_inputs)

Kind: instance method of HubertForCTC