transformers の API リファレンス

transformers (huggingface/transformers) の公開 API 400 件 —— クラス 229、関数 63、メソッド 108。実際のソースを静的解析して抽出した正確なシグネチャを掲載しています。

種別	件数
クラス	229
関数	63
メソッド	108

API 一覧

classbenchmark_v2.framework.benchmark_config.BenchmarkConfig

Configuration for a single benchmark scenario.

method

benchmark_v2.framework.benchmark_runner.BenchmarkRunner.save_results(model_name:str, results:dict, timestamp:str='', summarized:bool=True) -> str

Save benchmark results to JSON file.

funcbenchmark_v2.framework.benchmark_runner.flush_memory(flush_compile:bool=True) -> None

Flush GPU memory and run garbage collection.

classbenchmark_v2.framework.data_classes.BenchmarkMetadata

Metadata collected for each benchmark run.

classbenchmark_v2.framework.data_classes.BenchmarkResult

Result from a series of benchmark runs.

classbenchmark_v2.framework.hardware_metrics.GPUMonitoringStatus

Status of GPU monitoring.

classbenchmark_v2.framework.hardware_metrics.GPURawMetrics

Raw values for GPU utilization and memory used.

classbenchmark_v2.framework.hardware_metrics.HardwareInfo

A class to hold information about the hardware.

funcbenchmark_v2.framework.hardware_metrics.get_amd_gpu_stats(device_handle) -> tuple[int, float]

Get AMD GPU stats using amdsmi library.

classsrc.transformers.activations.LinearActivation

Applies the linear activation function, i.e.

funcsrc.transformers.audio_utils.get_audio_filetype(data:bytes) -> str

Identify a file's container/codec from its magic bytes.

funcsrc.transformers.audio_utils.hertz_to_mel(freq:float | np.ndarray, mel_scale:str='htk') -> float | np.ndarray

Convert frequency from hertz to mels.

func

src.transformers.audio_utils.load_audio(audio:str | np.ndarray, sampling_rate=16000, timeout=None, backend:str='auto') -> np.ndarray

Loads `audio` to an np.ndarray object.

funcsrc.transformers.audio_utils.load_audio_librosa(audio:str | np.ndarray, sampling_rate=16000, timeout=None) -> np.ndarray

Deprecated.

func

src.transformers.audio_utils.load_audio_torchcodec(audio:str | np.ndarray, sampling_rate=16000, timeout=None) -> np.ndarray

Deprecated.

funcsrc.transformers.audio_utils.make_list_of_audio(audio:list[AudioInput] | AudioInput) -> AudioInput

Ensure that the output is a list of audio.

func

src.transformers.audio_utils.make_list_of_audio_chat_template(audio:list[AudioInput] | AudioInput | str | list[str]) -> AudioInput

Ensure that the output is a list of audio.

funcsrc.transformers.audio_utils.mel_to_hertz(mels:float | np.ndarray, mel_scale:str='htk') -> float | np.ndarray

Convert frequency from mels to hertz.

funcsrc.transformers.audio_utils.optimal_fft_length(window_length:int) -> int

Finds the best FFT input size for a given `window_length`.

methodsrc.transformers.cache_utils.Cache.batch_repeat_interleave(repeats:int)

Repeat and interleave the cache

methodsrc.transformers.cache_utils.Cache.batch_select_indices(indices:torch.Tensor)

Select indices from the cache

methodsrc.transformers.cache_utils.Cache.crop(tokens_to_remove:int) -> None

Remove `tokens_to_remove` tokens from the current Cache.

methodsrc.transformers.cache_utils.Cache.get_max_length(layer_idx:int | None=None) -> int

Returns the maximum length of the cache.

methodsrc.transformers.cache_utils.Cache.get_seq_length(layer_idx:int=0) -> int

Returns the sequence length of the cache for the given layer.

methodsrc.transformers.cache_utils.Cache.is_compileable() -> bool

Return whether the cache is compilable

methodsrc.transformers.cache_utils.Cache.is_initialized() -> bool

Return whether the cache data is initialized

methodsrc.transformers.cache_utils.Cache.offload(layer_idx:int, only_non_sliding:bool=True)

Offload a given `layer_idx`.

methodsrc.transformers.cache_utils.Cache.reorder_cache(beam_idx:torch.LongTensor)

Reorder the cache for beam search

methodsrc.transformers.cache_utils.Cache.update_indexer(indexer_key_states:torch.Tensor, layer_idx:int) -> torch.Tensor

Updates the indexer key cache for layer `layer_idx`.

classsrc.transformers.cache_utils.CacheLayerMixin

Base, abstract class for a single layer's cache.

methodsrc.transformers.cache_utils.CacheLayerMixin.get_max_length() -> int

Returns the maximum sequence length the layer can hold.

methodsrc.transformers.cache_utils.CacheLayerMixin.reorder_cache(beam_idx:torch.LongTensor) -> None

Reorders this layer's cache for beam search.

methodsrc.transformers.cache_utils.CacheLayerMixin.reset() -> None

Resets the cache values while preserving the objects

classsrc.transformers.cache_utils.EncoderDecoderCache

Base, abstract class for all encoder-decoder caches.

methodsrc.transformers.cache_utils.EncoderDecoderCache.get_max_length(layer_idx:int | None=None) -> int

Returns the maximum sequence length (i.e.

methodsrc.transformers.cache_utils.EncoderDecoderCache.get_seq_length(layer_idx:int=0) -> int

Returns the sequence length of the cached states.

classsrc.transformers.cli.add_new_model_like.ClassFinder

A visitor to find all classes in a python module.

methodsrc.transformers.cli.add_new_model_like.ClassFinder.visit_ClassDef(node:cst.ClassDef) -> None

Record class names.

funcsrc.transformers.cli.add_new_model_like.convert_to_bool(x:str) -> bool

Converts a string to a bool.

classsrc.transformers.cli.chat.Chat

Chat with a model from the command line.

funcsrc.transformers.cli.chat.get_username() -> str

Returns the username of the current user.

funcsrc.transformers.cli.chat.new_chat_history(system_prompt:str | None=None) -> list[dict]

Returns a new chat conversation.

funcsrc.transformers.cli.chat.save_chat(filename:str, chat:list[dict], settings:dict) -> str

Saves the chat history to a file.

classsrc.transformers.cli.serving.completion.CompletionHandler

Handler for the `/v1/completions` endpoint.

methodsrc.transformers.cli.serving.model_manager.ModelManager.process_model_name(model_id:str) -> str

Canonicalize to `'model_id@revision'` format.

methodsrc.transformers.cli.serving.model_manager.ModelManager.shutdown() -> None

Delete all loaded models and free resources.

classsrc.transformers.cli.serving.response.ResponseHandler

Handler for the ``/v1/responses`` endpoint.

classsrc.transformers.cli.serving.transcription.TranscriptionHandler

Handler for ``POST /v1/audio/transcriptions``.

classsrc.transformers.cli.serving.utils.BaseGenerateManager

Base class for generation managers.

method

src.transformers.cli.serving.utils.BaseGenerateManager.init_cb(model:'PreTrainedModel', gen_config:'GenerationConfig') -> None

Initialize continuous batching.

methodsrc.transformers.cli.serving.utils.BaseGenerateManager.stop() -> None

Stop the generation manager and free resources.

classsrc.transformers.cli.serving.utils.BaseHandler

Shared logic for chat completion and responses handlers.

classsrc.transformers.cli.serving.utils.CBGenerateManager

Continuous batching generation via paged attention.

methodsrc.transformers.cli.serving.utils.CBGenerateManager.is_alive() -> bool

Whether the CB worker is healthy.

methodsrc.transformers.cli.serving.utils.CBGenerateManager.scheduler() -> 'Scheduler'

The CB scheduler (for testing/monitoring).

classsrc.transformers.cli.serving.utils.GenerationState

Shared generation state across all handlers.

methodsrc.transformers.cli.serving.utils.GenerationState.is_cb_alive() -> bool

Whether the CB worker is healthy.

methodsrc.transformers.cli.serving.utils.GenerationState.shutdown() -> None

Stop any active generation managers.

classsrc.transformers.cli.serving.utils.InferenceThread

Persistent thread for ``model.generate()`` calls.

methodsrc.transformers.cli.serving.utils.InferenceThread.submit(fn, *args, **kwargs) -> Future

Submit a callable to the inference thread.

funcsrc.transformers.cli.serving.utils.reset_torch_cache() -> None

Empty the CUDA cache if a GPU is available.

funcsrc.transformers.cli.serving.utils.set_torch_seed(seed:int) -> None

Set the PyTorch random seed for reproducible generation.

funcsrc.transformers.cli.system.version() -> None

Print CLI version.

classsrc.transformers.configuration_utils.PreTrainedConfig

Base class for all configuration classes.

methodsrc.transformers.configuration_utils.PreTrainedConfig.to_json_string(use_diff:bool=True) -> str

Serializes this instance to a JSON string.

funcsrc.transformers.configuration_utils.remap_legacy_layer_types(layer_types:list[str]) -> list[str]

Apply legacy → current layer-type name mapping.

classsrc.transformers.convert_slow_tokenizer.TikTokenConverter

A general tiktoken converter.

classsrc.transformers.core_model_loading.Chunk

Split a tensor along `dim` into equally sized chunks.

classsrc.transformers.core_model_loading.Concatenate

Concatenate tensors along `dim`.

classsrc.transformers.core_model_loading.Conv3dToLinear

Conv3d weights → flattened Linear layout.

classsrc.transformers.core_model_loading.ConversionOps

Base class for weight conversion operations.

classsrc.transformers.core_model_loading.LinearToConv3d

Flattened Linear weights → Conv3d layout.

classsrc.transformers.core_model_loading.Transpose

Transposes the given tensor along dim0 and dim1.

classsrc.transformers.data.data_collator.DataCollatorForLanguageModeling

Data collator used for language modeling.

classsrc.transformers.data.data_collator.DataCollatorWithFlattening

Data collator used for padding free approach.

classsrc.transformers.data.processors.glue.ColaProcessor

Processor for the CoLA data set (GLUE version).

methodsrc.transformers.data.processors.glue.ColaProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.ColaProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.ColaProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.ColaProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.ColaProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.MnliProcessor

Processor for the MultiNLI data set (GLUE version).

methodsrc.transformers.data.processors.glue.MnliProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.MnliProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.MnliProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.MnliProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.MnliProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.MrpcProcessor

Processor for the MRPC data set (GLUE version).

methodsrc.transformers.data.processors.glue.MrpcProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.MrpcProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.MrpcProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.MrpcProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.MrpcProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.QnliProcessor

Processor for the QNLI data set (GLUE version).

methodsrc.transformers.data.processors.glue.QnliProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.QnliProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.QnliProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.QnliProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.QnliProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.QqpProcessor

Processor for the QQP data set (GLUE version).

methodsrc.transformers.data.processors.glue.QqpProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.QqpProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.QqpProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.QqpProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.QqpProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.RteProcessor

Processor for the RTE data set (GLUE version).

methodsrc.transformers.data.processors.glue.RteProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.RteProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.RteProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.RteProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.RteProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.Sst2Processor

Processor for the SST-2 data set (GLUE version).

methodsrc.transformers.data.processors.glue.Sst2Processor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.Sst2Processor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.Sst2Processor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.Sst2Processor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.Sst2Processor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.StsbProcessor

Processor for the STS-B data set (GLUE version).

methodsrc.transformers.data.processors.glue.StsbProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.StsbProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.StsbProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.StsbProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.StsbProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.glue.WnliProcessor

Processor for the WNLI data set (GLUE version).

methodsrc.transformers.data.processors.glue.WnliProcessor.get_dev_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.WnliProcessor.get_example_from_tensor_dict(tensor_dict)

See base class.

methodsrc.transformers.data.processors.glue.WnliProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.glue.WnliProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.glue.WnliProcessor.get_train_examples(data_dir)

See base class.

classsrc.transformers.data.processors.squad.SquadFeatures

Single squad example features to be fed to a model.

classsrc.transformers.data.processors.squad.SquadProcessor

Processor for the SQuAD data set.

classsrc.transformers.data.processors.utils.InputExample

A single training/test example for simple sequence classification.

classsrc.transformers.data.processors.utils.InputFeatures

A single set of features of data.

classsrc.transformers.data.processors.xnli.XnliProcessor

Processor for the XNLI dataset.

methodsrc.transformers.data.processors.xnli.XnliProcessor.get_labels()

See base class.

methodsrc.transformers.data.processors.xnli.XnliProcessor.get_test_examples(data_dir)

See base class.

methodsrc.transformers.data.processors.xnli.XnliProcessor.get_train_examples(data_dir)

See base class.

func

src.transformers.distributed.fsdp.apply_fully_sharded_data_parallelism(model:nn.Module, fsdp_mesh:torch.distributed.device_mesh.DeviceMesh) -> nn.Module

Apply FSDP2 (fully_shard) to a model.

funcsrc.transformers.distributed.fsdp.is_fsdp_managed_module(module:nn.Module) -> bool

Check if a module is managed by FSDP (1 or 2).

funcsrc.transformers.distributed.utils.load_optimizer_distributed(model, optimizer, checkpoint_dir:str) -> None

Load optimizer state via DCP.

funcsrc.transformers.distributed.utils.save_optimizer_distributed(model, optimizer, checkpoint_dir:str) -> None

Save optimizer state via DCP.

classsrc.transformers.exporters.base.HfExporter

Abstract base class for all Transformers exporters.

classsrc.transformers.exporters.configs.ExportFormat

Identifies the export backend.

funcsrc.transformers.exporters.exporter_dynamo.get_auto_dynamic_shapes(inputs:Any) -> Any

Recursively build dynamic shapes for any input value.

funcsrc.transformers.exporters.exporter_dynamo.register_pytree_node(object_cls:type)

funcsrc.transformers.exporters.exporter_executorch.prepare_for_xnnpack(model:PreTrainedModel, sample_inputs:dict[str, Any])

CPU inference via XNNPACK.

funcsrc.transformers.exporters.utils.module_device(model:PreTrainedModel | torch.nn.Module) -> torch.device | None

`.device` for any `nn.Module`.

funcsrc.transformers.exporters.utils.module_dtype(model:PreTrainedModel | torch.nn.Module) -> torch.dtype | None

`.dtype` for any `nn.Module`.

funcsrc.transformers.exporters.utils.register_export_input_preparer(*markers:str)

classsrc.transformers.fusion_mapping.ModuleFusionSpec

Base recipe for a fusion family.

classsrc.transformers.generation.configuration_utils.BaseWatermarkingConfig

Generic watermarking config

method

src.transformers.generation.configuration_utils.GenerationConfig.to_json_file(json_file_path:str | os.PathLike, use_diff:bool=True, keys_to_pop:list[str] | None=None) -> None

Save this instance to a JSON file.

funcsrc.transformers.generation.continuous_batching.cache.find_head_dim(config:PreTrainedConfig) -> int

Finds the head dimension for the given config.

method

src.transformers.generation.continuous_batching.cache_manager.BlockManager.free_blocks(blocks:list[int], shareable:bool) -> None

Marks a list of (blocks) as free.

method

src.transformers.generation.continuous_batching.cache_manager.BlockManager.uninitialize_unshared_block(block_id:int) -> None

Marks a block as uninitialized.

classsrc.transformers.generation.continuous_batching.cache_manager.CacheAllocator

Abstract base class for cache managers.

classsrc.transformers.generation.continuous_batching.requests.GenerationOutput

Tracks the output of a generation request.

classsrc.transformers.generation.logits_process.MinPLogitsWarper

[`LogitsProcessor`] that performs min-p, i.e.

classsrc.transformers.generation.logits_process.SynthIDTextWatermarkState

SynthID watermarking state.

classsrc.transformers.generation.logits_process.TopKLogitsWarper

[`LogitsProcessor`] that performs top-k, i.e.

classsrc.transformers.generation.logits_process.TopPLogitsWarper

[`LogitsProcessor`] that performs top-p, i.e.

classsrc.transformers.generation.streamers.TextDiffusionStreamer

Streamer that prints text diffusion outputs.

methodsrc.transformers.generation.streamers.TextStreamer.on_finalized_text(text:str, stream_end:bool=False)

Prints the new text to stdout.

classsrc.transformers.generation.watermarking.BayesianDetectorModel

Bayesian classifier for watermark detection.

classsrc.transformers.generation.watermarking.SynthIDTextWatermarkDetector

SynthID text watermark detector class.

classsrc.transformers.generation.watermarking.WatermarkDetectorOutput

Outputs of a watermark detector.

method

src.transformers.image_processing_backends.PilBackend.center_crop(image:np.ndarray, size:SizeDict, **kwargs) -> np.ndarray

Center crop an image using NumPy.

methodsrc.transformers.image_processing_backends.PilBackend.convert_to_rgb(image:ImageInput) -> ImageInput

Convert an image to RGB format.

method

src.transformers.image_processing_backends.PilBackend.normalize(image:np.ndarray, mean:float | Iterable[float], std:float | Iterable[float], **kwargs) -> np.ndarray

Normalize an image using NumPy.

methodsrc.transformers.image_processing_backends.PilBackend.rescale(image:np.ndarray, scale:float, **kwargs) -> np.ndarray

Rescale an image by a scale factor using NumPy.

method

src.transformers.image_processing_backends.PilBackend.resize(image:np.ndarray, size:SizeDict, resample:'PILImageResampling | None'=None, reducing_gap:int | None=None, **kwargs) -> np.ndarray

Resize an image using PIL/NumPy.

funcsrc.transformers.image_transforms.convert_to_rgb(image:ImageInput) -> ImageInput

Converts an image to RGB format.

func

src.transformers.image_transforms.group_images_by_shape(images:Union[list['torch.Tensor'], 'torch.Tensor'], *disable_grouping:bool | None, *is_nested:bool=False, *paired_inputs) -> tuple[dict, ...]

Groups images by shape.

funcsrc.transformers.image_transforms.id_to_rgb(id_map)

Converts unique ID to RGB color.

func

src.transformers.image_transforms.rescale(image:np.ndarray, scale:float, data_format:ChannelDimension | None=None, dtype:np.dtype=np.float32, input_data_format:str | ChannelDimension | None=None) -> np.ndarray

Rescales `image` by `scale`.

funcsrc.transformers.image_transforms.rgb_to_id(color)

Converts RGB color to unique ID.

classsrc.transformers.image_utils.SizeDict

Hashable dictionary to store image size information.

func

src.transformers.image_utils.infer_channel_dimension_format(image:np.ndarray, num_channels:int | tuple[int, ...] | None=None) -> ChannelDimension

Infers the channel dimension format of `image`.

func

src.transformers.image_utils.load_image(image:Union[str, 'PIL.Image.Image'], timeout:float | None=None) -> 'PIL.Image.Image'

Loads `image` to a PIL Image.

func

src.transformers.image_utils.make_flat_list_of_images(images:list[ImageInput] | ImageInput, expected_ndims:int=3) -> ImageInput

Ensure that the output is a flat list of images.

funcsrc.transformers.image_utils.make_list_of_images(images, expected_ndims:int=3) -> list[ImageInput]

Ensure that the output is a list of images.

classsrc.transformers.integrations.finegrained_fp8.FP8GroupedLinear

FP8 drop-in for block-diagonal grouped linears.

classsrc.transformers.integrations.integration_utils.WandbLogModel

Enum of possible log model values in W&B.

funcsrc.transformers.integrations.integration_utils.default_logdir() -> str

Same default as PyTorch

classsrc.transformers.integrations.sinq.SinqQuantize

Param-level ConversionOp for SINQ (from FP weights).

classsrc.transformers.integrations.tensor_parallel.MoeTensorParalellMegaMoeExperts

TP layer for DeepGEMM Mega MoE experts.

classsrc.transformers.integrations.tensor_parallel.RouterParallelMegaMoe

Router TP plan used with DeepGEMM Mega MoE.

classsrc.transformers.integrations.tensor_parallel.TensorParallelLayer

General tensor parallel layer for transformers

funcsrc.transformers.integrations.tensor_parallel.split(x, device_mesh)

Split forward, all-gather backward.

classsrc.transformers.loss.loss_d_fine.DFineLoss

This class computes the losses for D-FINE.

classsrc.transformers.loss.loss_rt_detr.RTDetrLoss

This class computes the losses for RTDetr.

funcsrc.transformers.masking_utils.bidirectional_mask_function(batch_idx:int, head_idx:int, q_idx:int, kv_idx:int) -> bool

This creates a full bidirectional mask.

funcsrc.transformers.masking_utils.causal_mask_function(batch_idx:int, head_idx:int, q_idx:int, kv_idx:int) -> bool

This creates a basic lower-diagonal causal mask.

funcsrc.transformers.masking_utils.sliding_window_overlay(sliding_window:int) -> Callable

This is an overlay depicting a sliding window pattern.

classsrc.transformers.modeling_layers.GradientCheckpointingLayer

Base class for layers with gradient checkpointing.

classsrc.transformers.modeling_outputs.BackboneOutput

Base class for outputs of backbones.

classsrc.transformers.modeling_outputs.DepthEstimatorOutput

Base class for outputs of depth estimation models.

classsrc.transformers.modeling_outputs.MaskedLMOutput

Base class for masked language models outputs.

classsrc.transformers.modeling_outputs.MultipleChoiceModelOutput

Base class for outputs of multiple choice models.

classsrc.transformers.modeling_outputs.XVectorOutput

Output type of [`Wav2Vec2ForXVector`].

classsrc.transformers.modeling_utils.LoadStateDictConfig

Config for loading weights.

classsrc.transformers.modeling_utils.PreTrainedModel

Base class for all models.

methodsrc.transformers.modeling_utils.PreTrainedModel.base_model() -> nn.Module

`torch.nn.Module`: The main body of the model.

methodsrc.transformers.modeling_utils.PreTrainedModel.set_decoder(decoder)

Symmetric setter.

methodsrc.transformers.modeling_utils.PreTrainedModel.set_encoder(encoder, modality:str | None=None)

Symmetric setter.

method

src.transformers.modeling_utils.PreTrainedModel.tie_weights(missing_keys:set[str] | None=None, recompute_mapping:bool=True)

Tie the model weights.

funcsrc.transformers.modeling_utils.is_accelerator_device(device:str | int | torch.device) -> bool

Check if the device is an accelerator.

classsrc.transformers.models.afmoe.modeling_afmoe.AfmoeDecoderLayer

AFMoE decoder layer with dual normalization.

classsrc.transformers.models.afmoe.modeling_afmoe.AfmoeExperts

Collection of expert weights stored as 3D tensors.

classsrc.transformers.models.afmoe.modeling_afmoe.AfmoeSparseMoeBlock

Mixture of Experts (MoE) module for AFMoE.

classsrc.transformers.models.afmoe.modeling_afmoe.AfmoeTokenChoiceRouter

Token-choice top-K router for MoE routing.

classsrc.transformers.models.afmoe.modular_afmoe.AfmoeDecoderLayer

AFMoE decoder layer with dual normalization.

classsrc.transformers.models.afmoe.modular_afmoe.AfmoeSparseMoeBlock

Mixture of Experts (MoE) module for AFMoE.

classsrc.transformers.models.afmoe.modular_afmoe.AfmoeTokenChoiceRouter

Token-choice top-K router for MoE routing.

classsrc.transformers.models.aria.modeling_aria.AriaCrossAttention

Aria Cross-Attention module.

classsrc.transformers.models.aria.modeling_aria.AriaProjector

Aria Projector module.

classsrc.transformers.models.aria.modeling_aria.AriaProjectorMLP

Feed-Forward Network module for the Aria Projector.

classsrc.transformers.models.aria.modeling_aria.AriaSharedExpertsMLP

Shared Expert MLP for shared experts.

classsrc.transformers.models.aria.modeling_aria.AriaTextDecoderLayer

Aria Text Decoder Layer.

classsrc.transformers.models.aria.modular_aria.AriaCrossAttention

Aria Cross-Attention module.

classsrc.transformers.models.aria.modular_aria.AriaProjector

Aria Projector module.

classsrc.transformers.models.aria.modular_aria.AriaSharedExpertsMLP

Shared Expert MLP for shared experts.

classsrc.transformers.models.aria.modular_aria.AriaTextDecoderLayer

Aria Text Decoder Layer.

classsrc.transformers.models.autoformer.modeling_autoformer.AutoformerFeatureEmbedder

Embed a sequence of categorical features.

classsrc.transformers.models.axk1.modeling_axk1.AXK1Experts

Collection of expert weights stored as 3D tensors.

classsrc.transformers.models.axk2.modeling_axk2.AXK2Experts

Collection of expert weights stored as 3D tensors.

classsrc.transformers.models.axk2.modeling_axk2.AXK2MoE

A mixed expert module containing shared experts.

classsrc.transformers.models.bark.modeling_bark.BarkSelfFlashAttention2

Bark flash attention module.

classsrc.transformers.models.bart.modeling_bart.BartClassificationHead

Head for sentence-level classification tasks.

classsrc.transformers.models.bartpho.tokenization_bartpho.BartphoTokenizer

Adapted from [`XLMRobertaTokenizer`].

classsrc.transformers.models.beit.modeling_beit.BeitFPNNeck

4-level feature pyramid neck for BeiT.

classsrc.transformers.models.beit.modeling_beit.BeitPyramidPoolingModule

Pyramid Pooling Module (PPM) used in PSPNet.

classsrc.transformers.models.beit.modular_beit.BeitFPNNeck

4-level feature pyramid neck for BeiT.

classsrc.transformers.models.beit.modular_beit.BeitPyramidPoolingModule

Pyramid Pooling Module (PPM) used in PSPNet.

classsrc.transformers.models.beit.modular_beit.BeitUperHead

Unified Perceptual Parsing for Scene Understanding.

classsrc.transformers.models.bert.tokenization_bert_legacy.BertTokenizerLegacy

Construct a BERT tokenizer.

classsrc.transformers.models.bert.tokenization_bert_legacy.WordpieceTokenizer

Runs WordPiece tokenization.

classsrc.transformers.models.bert_generation.tokenization_bert_generation.BertGenerationTokenizer

Construct a BertGeneration tokenizer.

classsrc.transformers.models.bert_japanese.tokenization_bert_japanese.CharacterTokenizer

Runs Character tokenization.

classsrc.transformers.models.bert_japanese.tokenization_bert_japanese.SentencepieceTokenizer

Runs sentencepiece tokenization.

classsrc.transformers.models.bert_japanese.tokenization_bert_japanese.WordpieceTokenizer

Runs WordPiece tokenization.

classsrc.transformers.models.big_bird.modeling_big_bird.BigBirdForQuestionAnsweringHead

Head for question answering tasks.

classsrc.transformers.models.biogpt.tokenization_biogpt.BioGptTokenizer

Construct an FAIRSEQ Transformer tokenizer.

methodsrc.transformers.models.biogpt.tokenization_biogpt.BioGptTokenizer.vocab_size()

Returns vocab size

classsrc.transformers.models.bit.modeling_bit.BitPreActivationBottleneckLayer

Pre-activation (v2) bottleneck block.

classsrc.transformers.models.bit.modeling_bit.BitStage

A ResNet v2 stage composed by stacked layers.

classsrc.transformers.models.bit.modeling_bit.WeightStandardizedConv2d

Conv2d with Weight Standardization.

funcsrc.transformers.models.bloom.modeling_bloom.bloom_gelu_forward(x:torch.Tensor) -> torch.Tensor

Custom bias GELU function.

classsrc.transformers.models.byt5.tokenization_byt5.ByT5Tokenizer

Construct a ByT5 tokenizer.

classsrc.transformers.models.camembert.modeling_camembert.CamembertLMHead

Camembert Head for masked language modeling.

classsrc.transformers.models.canine.tokenization_canine.CanineTokenizer

Construct a CANINE tokenizer (i.e.

func

src.transformers.models.chmv2.convert_chmv2_to_hf.get_chmv2_config(model_name:str, backbone_repo_id:str | None=None) -> CHMv2Config

Create CHMv2 config based on model name.

classsrc.transformers.models.chmv2.modeling_chmv2.CHMv2Head

CHMv2 dense-prediction head adapted from DPT.

classsrc.transformers.models.chmv2.modeling_chmv2.CHMv2PreActResidualLayer

ResidualConvUnit, pre-activate residual unit.

classsrc.transformers.models.chmv2.modeling_chmv2.CHMv2UpsampleConvHead

Convolutional head with intermediate upsampling.

classsrc.transformers.models.chmv2.modular_chmv2.CHMv2Head

CHMv2 dense-prediction head adapted from DPT.

classsrc.transformers.models.clap.feature_extraction_clap.ClapFeatureExtractor

Constructs a CLAP feature extractor.

classsrc.transformers.models.clap.modeling_clap.ClapAudioPatchMerging

Patch Merging Layer.

funcsrc.transformers.models.clap.modeling_clap.interpolate(hidden_states, ratio)

Interpolate data in time domain.

classsrc.transformers.models.clvp.feature_extraction_clvp.ClvpFeatureExtractor

Constructs a CLVP feature extractor.

classsrc.transformers.models.clvp.modeling_clvp.ClvpRotaryPositionalEmbedding

Rotary Position Embedding Class for CLVP.

classsrc.transformers.models.clvp.tokenization_clvp.ClvpTokenizer

Construct a CLVP tokenizer.

classsrc.transformers.models.code_llama.tokenization_code_llama.CodeLlamaTokenizer

Construct a Llama tokenizer.

classsrc.transformers.models.cohere.tokenization_cohere.CohereTokenizer

Construct a Cohere tokenizer.

classsrc.transformers.models.convnextv2.modeling_convnextv2.ConvNextV2GRN

GRN (Global Response Normalization) layer

classsrc.transformers.models.cpmant.tokenization_cpmant.CpmAntTokenizer

Construct a CPMAnt tokenizer.

classsrc.transformers.models.csm.generation_csm.CsmGenerateOutput

Outputs of CsmForConditionalGeneration.generate.

classsrc.transformers.models.ctrl.tokenization_ctrl.CTRLTokenizer

Construct a CTRL tokenizer.

classsrc.transformers.models.cvt.modeling_cvt.CvtConvEmbeddings

Image to Conv Embedding.

classsrc.transformers.models.cvt.modeling_cvt.CvtEmbeddings

Construct the CvT embeddings.

classsrc.transformers.models.dac.feature_extraction_dac.DacFeatureExtractor

Constructs an Dac feature extractor.

classsrc.transformers.models.dac.modeling_dac.DacDecoder

DAC Decoder

classsrc.transformers.models.dac.modeling_dac.DacDecoderBlock

Decoder block used in DAC decoder.

classsrc.transformers.models.dac.modeling_dac.DacEncoder

DAC Encoder

classsrc.transformers.models.dac.modeling_dac.DacEncoderBlock

Encoder block used in DAC encoder.

classsrc.transformers.models.dac.modeling_dac.Snake1d

A 1-dimensional Snake activation function module.

classsrc.transformers.models.dbrx.modeling_dbrx.DbrxFFN

Modular DBRX MLP/FFN component with MoE support.

classsrc.transformers.models.dbrx.modular_dbrx.DbrxFFN

Modular DBRX MLP/FFN component with MoE support.

classsrc.transformers.models.deepseek_ocr2.modeling_deepseek_ocr2.DeepseekOcr2VisionModel

Vision pipeline: SAM ViT-B (with neck)

classsrc.transformers.models.deepseek_ocr2.modular_deepseek_ocr2.DeepseekOcr2VisionModel

Vision pipeline: SAM ViT-B (with neck)

classsrc.transformers.models.deepseek_v4.modeling_deepseek_v4.DeepseekV4CSACache

Cache layer for CSA blocks (paper §2.3.1).

classsrc.transformers.models.deepseek_v4.modeling_deepseek_v4.DeepseekV4DecoderLayer

DeepSeek-V4 decoder block (paper §2).

classsrc.transformers.models.deepseek_v4.modeling_deepseek_v4.DeepseekV4HCACache

Cache layer for HCA blocks (paper §2.3.2).

classsrc.transformers.models.deepseek_v4.modeling_deepseek_v4.DeepseekV4Indexer

Lightning Indexer (paper §2.3.1, eqs.

classsrc.transformers.models.deepseek_v4.modular_deepseek_v4.DeepseekV4CSACache

Cache layer for CSA blocks (paper §2.3.1).

classsrc.transformers.models.deepseek_v4.modular_deepseek_v4.DeepseekV4DecoderLayer

DeepSeek-V4 decoder block (paper §2).

classsrc.transformers.models.deepseek_v4.modular_deepseek_v4.DeepseekV4HCACache

Cache layer for HCA blocks (paper §2.3.2).

classsrc.transformers.models.deepseek_v4.modular_deepseek_v4.DeepseekV4Indexer

Lightning Indexer (paper §2.3.1, eqs.

classsrc.transformers.models.deimv2.modeling_deimv2.Deimv2HybridEncoder

DEIMv2 variant of DFineHybridEncoder.

classsrc.transformers.models.deimv2.modular_deimv2.Deimv2HybridEncoder

DEIMv2 variant of DFineHybridEncoder.

classsrc.transformers.models.depth_anything.modeling_depth_anything.DepthAnythingNeck

DepthAnythingNeck.

func

src.transformers.models.depth_pro.modeling_depth_pro.split_to_patches(pixel_values:torch.Tensor, patch_size:int, overlap_ratio:float) -> torch.Tensor

Creates Patches from Batch.

func

src.transformers.models.detr.image_processing_pil_detr.convert_coco_poly_to_mask(segmentations, height:int, width:int) -> np.ndarray

Convert a COCO polygon annotation to a mask.

classsrc.transformers.models.dia.feature_extraction_dia.DiaFeatureExtractor

Constructs an Dia feature extractor.

classsrc.transformers.models.dia.modeling_dia.DiaDecoder

Transformer Decoder Stack using DenseGeneral.

classsrc.transformers.models.dia.modular_dia.DiaDecoder

Transformer Decoder Stack using DenseGeneral.

classsrc.transformers.models.dia.tokenization_dia.DiaTokenizer

Construct a Dia tokenizer.

classsrc.transformers.models.diffusion_gemma.modeling_diffusion_gemma.DiffusionGemmaDecoderModel

Decoder model for DiffusionGemma.

classsrc.transformers.models.diffusion_gemma.modular_diffusion_gemma.DiffusionGemmaDecoderModel

Decoder model for DiffusionGemma.

classsrc.transformers.models.dinat.modeling_dinat.DinatDownsampler

Convolutional Downsampling Layer.

classsrc.transformers.models.dinat.modeling_dinat.DinatEmbeddings

Construct the patch and position embeddings.

classsrc.transformers.models.donut.modeling_donut_swin.DonutSwinEmbeddings

Construct the patch and position embeddings.

classsrc.transformers.models.donut.modeling_donut_swin.DonutSwinPatchMerging

Patch Merging Layer.

classsrc.transformers.models.dots1.modeling_dots1.Dots1Experts

Collection of expert weights stored as 3D tensors.

classsrc.transformers.models.dots1.modeling_dots1.Dots1MoE

A mixed expert module containing shared experts.

classsrc.transformers.models.dpr.tokenization_dpr.DPRContextEncoderTokenizer

Construct a DPRContextEncoder tokenizer.

classsrc.transformers.models.dpr.tokenization_dpr.DPRQuestionEncoderTokenizer

Constructs a DPRQuestionEncoder tokenizer.

classsrc.transformers.models.dpr.tokenization_dpr.DPRReaderTokenizer

Construct a DPRReader tokenizer.

classsrc.transformers.models.dpt.image_processing_dpt.DPTImageProcessor

PIL backend for DPT with reduce_label support.

classsrc.transformers.models.dpt.modeling_dpt.DPTNeck

DPTNeck.

classsrc.transformers.models.dpt.modeling_dpt.DPTPreActResidualLayer

ResidualConvUnit, pre-activate residual unit.

classsrc.transformers.models.dpt.modeling_dpt.DPTViTPatchEmbeddings

Image to Patch Embedding.

classsrc.transformers.models.encodec.feature_extraction_encodec.EncodecFeatureExtractor

Constructs an EnCodec feature extractor.

classsrc.transformers.models.encodec.modeling_encodec.EncodecDecoder

SEANet decoder as used by EnCodec.

classsrc.transformers.models.encodec.modeling_encodec.EncodecEncoder

SEANet encoder as used by EnCodec.

classsrc.transformers.models.encodec.modeling_encodec.EncodecEuclideanCodebook

Codebook with Euclidean distance.

classsrc.transformers.models.encodec.modeling_encodec.EncodecResidualVectorQuantizer

Residual Vector Quantizer.

classsrc.transformers.models.encodec.modeling_encodec.EncodecVectorQuantization

Vector quantization implementation.

funcsrc.transformers.models.eomt.image_processing_eomt.get_target_size(size_dict:dict[str, int]) -> tuple[int, int]

Returns the height and width from a size dict.

funcsrc.transformers.models.eomt.image_processing_pil_eomt.get_target_size(size_dict:dict[str, int]) -> tuple[int, int]

Returns the height and width from a size dict.

classsrc.transformers.models.esm.modeling_esm.EsmClassificationHead

Head for sentence-level classification tasks.

classsrc.transformers.models.esm.modeling_esm.EsmLMHead

ESM Head for masked language modeling.

classsrc.transformers.models.esm.modeling_esm.EsmRotaryEmbedding

Rotary position embeddings.

classsrc.transformers.models.esm.modeling_esmfold.EsmFoldAngleResnet

Implements Algorithm 20, lines 11-14

classsrc.transformers.models.esm.modeling_esmfold.EsmFoldBackboneUpdate

Implements part of Algorithm 23.

classsrc.transformers.models.esm.modeling_esmfold.EsmFoldInvariantPointAttention

Implements Algorithm 22.

classsrc.transformers.models.esm.modeling_esmfold.EsmFoldTriangleMultiplicativeUpdate

Implements Algorithms 11 and 12.

classsrc.transformers.models.esm.modeling_esmfold.EsmForProteinFoldingOutput

frames (`torch.FloatTensor`): Output frames.

classsrc.transformers.models.esm.openfold_utils.protein.Protein

Protein structure representation.

funcsrc.transformers.models.esm.openfold_utils.protein.add_pdb_headers(prot:Protein, pdb_str:str) -> str

Add pdb headers to an existing PDB string.

funcsrc.transformers.models.esm.openfold_utils.protein.ideal_atom_mask(prot:Protein) -> np.ndarray

Computes an ideal atom mask.

funcsrc.transformers.models.esm.openfold_utils.protein.to_pdb(prot:Protein) -> str

Converts a `Protein` instance to a PDB string.

funcsrc.transformers.models.esm.openfold_utils.residue_constants.make_bond_key(atom1_name:str, atom2_name:str) -> str

Unique key to lookup bonds.

classsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid

A class representing a rigid transformation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.apply(pts:torch.Tensor) -> torch.Tensor

Applies the transformation to a coordinate tensor.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.cat(ts:Sequence[Rigid], dim:int) -> Rigid

Concatenates transformations along a new dimension.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.compose(r:Rigid) -> Rigid

Composes the current rigid object with another.

method

src.transformers.models.esm.openfold_utils.rigid_utils.Rigid.from_3_points(p_neg_x_axis:torch.Tensor, origin:torch.Tensor, p_xy_plane:torch.Tensor, eps:float=1e-08) -> Rigid

Implements algorithm 21.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.get_rots() -> Rotation

Getter for the rotation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.get_trans() -> torch.Tensor

Getter for the translation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.invert() -> Rigid

Inverts the transformation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.scale_translation(trans_scale_factor:float) -> Rigid

Scales the translation by a constant factor.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rigid.unsqueeze(dim:int) -> Rigid

Analogous to torch.unsqueeze.

classsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation

A 3D rotation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.apply(pts:torch.Tensor) -> torch.Tensor

Apply the current Rotation as a rotation matrix to a set of 3D coordinates.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.cat(rs:Sequence[Rotation], dim:int) -> Rotation

Concatenates rotations along one of the batch dimensions.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.get_rot_mats() -> torch.Tensor

Returns the underlying rotation as a rotation matrix tensor.

method

src.transformers.models.esm.openfold_utils.rigid_utils.Rotation.identity(shape, dtype:torch.dtype | None=None, device:torch.device | None=None, requires_grad:bool=True, fmt:str='quat') -> Rotation

Returns an identity Rotation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.invert() -> Rotation

Returns the inverse of the current Rotation.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.invert_apply(pts:torch.Tensor) -> torch.Tensor

The inverse of the apply() method.

methodsrc.transformers.models.esm.openfold_utils.rigid_utils.Rotation.unsqueeze(dim:int) -> Rotation

Analogous to torch.unsqueeze.

func

src.transformers.models.esm.openfold_utils.rigid_utils.quat_multiply(quat1:torch.Tensor, quat2:torch.Tensor) -> torch.Tensor

Multiply a quaternion by another quaternion.

func

src.transformers.models.esm.openfold_utils.rigid_utils.quat_multiply_by_vec(quat:torch.Tensor, vec:torch.Tensor) -> torch.Tensor

Multiply a quaternion by a pure-vector quaternion.

funcsrc.transformers.models.esm.openfold_utils.rigid_utils.quat_to_rot(quat:torch.Tensor) -> torch.Tensor

Converts a quaternion to a rotation matrix.

funcsrc.transformers.models.esm.openfold_utils.rigid_utils.rot_matmul(a:torch.Tensor, b:torch.Tensor) -> torch.Tensor

Performs matrix multiplication of two rotation matrix tensors.

funcsrc.transformers.models.esm.openfold_utils.rigid_utils.rot_vec_mul(r:torch.Tensor, t:torch.Tensor) -> torch.Tensor

Applies a rotation to a vector.

classsrc.transformers.models.esm.tokenization_esm.EsmTokenizer

Constructs an ESM tokenizer.

classsrc.transformers.models.evolla.modeling_evolla.EvollaSaProtRotaryEmbedding

Rotary position embeddings.

classsrc.transformers.models.falcon.modeling_falcon.FalconFlashAttention2

Falcon flash attention module.

classsrc.transformers.models.fastspeech2_conformer.modeling_fastspeech2_conformer.FastSpeech2ConformerDurationPredictor

Duration predictor module.

classsrc.transformers.models.fastspeech2_conformer.modeling_fastspeech2_conformer.FastSpeech2ConformerModel

FastSpeech 2 module.

classsrc.transformers.models.flaubert.modeling_flaubert.FlaubertSQuADHead

A SQuAD head inspired by XLNet.

classsrc.transformers.models.flaubert.tokenization_flaubert.FlaubertTokenizer

Construct a Flaubert tokenizer.

classsrc.transformers.models.flava.modeling_flava.PatchEmbeddings

Image to Patch Embedding.

classsrc.transformers.models.florence2.modeling_florence2.Florence2VisionConvEmbed

Image to Patch Embedding

classsrc.transformers.models.florence2.modular_florence2.Florence2PostProcessor

Post-processor for Florence-2 model outputs.

method

src.transformers.models.florence2.modular_florence2.Florence2PostProcessor.quantize(locations:'torch.Tensor', size:tuple[int, int]) -> 'torch.Tensor'

Quantize locations.

classsrc.transformers.models.florence2.modular_florence2.Florence2VisionConvEmbed

Image to Patch Embedding

method

src.transformers.models.florence2.processing_florence2.Florence2PostProcessor.quantize(locations:'torch.Tensor', size:tuple[int, int]) -> 'torch.Tensor'

Quantize locations.

classsrc.transformers.models.fnet.tokenization_fnet.FNetTokenizer

Construct an FNet tokenizer.

classsrc.transformers.models.focalnet.modeling_focalnet.FocalNetEmbeddings

Construct the patch embeddings and layernorm.

classsrc.transformers.models.focalnet.modeling_focalnet.FocalNetLayer

Focal Modulation Network layer (block).

funcsrc.transformers.models.fsmt.modeling_fsmt.make_padding_mask(input_ids, padding_idx=1)

True for pad tokens

classsrc.transformers.models.fsmt.tokenization_fsmt.FSMTTokenizer

Construct an FAIRSEQ Transformer tokenizer.

classsrc.transformers.models.gemma3n.modeling_gemma3n.Gemma3nTextLaurelBlock

Learned Augmented Residual Layer

classsrc.transformers.models.gemma3n.modular_gemma3n.Gemma3nTextLaurelBlock

Learned Augmented Residual Layer

funcsrc.transformers.models.gemma4.convert_gemma4_weights.store(path:str, weights, dtype=None)

Store a tensor in the HF tree.

classsrc.transformers.models.gemma4.modeling_gemma4.Gemma4VisionModel

The Gemma 4 Vision Encoder.

classsrc.transformers.models.gemma4.modular_gemma4.Gemma4VisionModel

The Gemma 4 Vision Encoder.

classsrc.transformers.models.gemma4_unified.modeling_gemma4_unified.Gemma4UnifiedModel

Encoder-free multimodal model.

classsrc.transformers.models.gemma4_unified.modular_gemma4_unified.Gemma4UnifiedModel

Encoder-free multimodal model.

classsrc.transformers.models.glm4_moe.modeling_glm4_moe.Glm4MoeMoE

A mixed expert module containing shared experts.

classsrc.transformers.models.glpn.modeling_glpn.GLPNOverlapPatchEmbeddings

Construct the overlapping patch embeddings.

classsrc.transformers.models.gpt2.tokenization_gpt2.GPT2Tokenizer

Construct a GPT-2 tokenizer.

classsrc.transformers.models.gpt_neo.modeling_gpt_neo.GPTNeoFlashAttention2

GPTNeo flash attention module.

classsrc.transformers.models.gpt_sw3.tokenization_gpt_sw3.GPTSw3Tokenizer

Construct an GPTSw3 tokenizer.

methodsrc.transformers.models.gpt_sw3.tokenization_gpt_sw3.GPTSw3Tokenizer.preprocess_text(text:str) -> str

Returns the preprocessed text.

classsrc.transformers.models.gptj.modeling_gptj.GPTJFlashAttention2

GPTJ flash attention module.

classsrc.transformers.models.groupvit.modeling_groupvit.GroupViTPatchEmbeddings

Image to Patch Embedding.

classsrc.transformers.models.hiera.modeling_hiera.HieraEmbeddings

Construct position and patch embeddings.

classsrc.transformers.models.hiera.modeling_hiera.HieraMaskUnitAttention

Computes either Mask Unit or Global Attention.

classsrc.transformers.models.hubert.modeling_hubert.HubertFeatureEncoder

Construct the features from raw audio waveform

classsrc.transformers.models.hy_v3.modeling_hy_v3.HYV3Experts

Collection of expert weights stored as 3D tensors.

classsrc.transformers.models.ibert.modeling_ibert.IBertClassificationHead

Head for sentence-level classification tasks.

classsrc.transformers.models.ibert.modeling_ibert.IBertLMHead

I-BERT Head for masked language modeling.

classsrc.transformers.models.ibert.quant_modules.IntGELU

Quantized version of `torch.nn.GELU`.

classsrc.transformers.models.ibert.quant_modules.IntLayerNorm

Quantized version of `torch.nn.LayerNorm`.

classsrc.transformers.models.ibert.quant_modules.IntSoftmax

Quantized version of `torch.nn.Softmax`.

classsrc.transformers.models.ibert.quant_modules.QuantAct

Quantizes the given activation.

classsrc.transformers.models.ibert.quant_modules.QuantEmbedding

Quantized version of `torch.nn.Embedding`.

この情報について

掲載しているシグネチャは huggingface/transformers の公開ソースコードを Python の ast モジュールで静的解析し、引数名・デフォルト値・型注釈・戻り値型をそのまま抽出したものです。実装コードは保存していません。詳しくは仕組みの解説をご覧ください。

収録ライブラリ一覧（全 805 件）へ戻る