# Exporting 🤗 Transformers models to ONNX 🤗 Transformers provides a `transformers.onnx` package that enables you to convert model checkpoints to an ONNX graph by leveraging configuration objects. See the [guide](../serialization) on exporting 🤗 Transformers models for more details. ## ONNX Configurations We provide three abstract classes that you should inherit from, depending on the type of model architecture you wish to export: * Encoder-based models inherit from [OnnxConfig](/docs/transformers/v4.57.0/en/main_classes/onnx#transformers.onnx.OnnxConfig) * Decoder-based models inherit from [OnnxConfigWithPast](/docs/transformers/v4.57.0/en/main_classes/onnx#transformers.onnx.OnnxConfigWithPast) * Encoder-decoder models inherit from [OnnxSeq2SeqConfigWithPast](/docs/transformers/v4.57.0/en/main_classes/onnx#transformers.onnx.OnnxSeq2SeqConfigWithPast) ### OnnxConfig[[transformers.onnx.OnnxConfig]]
class transformers.onnx.OnnxConfigtransformers.onnx.OnnxConfighttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L69[{"name": "config", "val": ": PretrainedConfig"}, {"name": "task", "val": ": str = 'default'"}, {"name": "patching_specs", "val": ": typing.Optional[list[transformers.onnx.config.PatchingSpec]] = None"}] Base class for ONNX exportable model describing metadata on how to export the model through the ONNX format.
flatten_output_collection_propertytransformers.onnx.OnnxConfig.flatten_output_collection_propertyhttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L427[{"name": "name", "val": ": str"}, {"name": "field", "val": ": Iterable"}]- **name** -- The name of the nested structure - **field** -- The structure to, potentially, be flattened0(dict[str, Any])Outputs with flattened structure and key mapping this new structure. Flatten any potential nested structure expanding the name of the field with the index of the element within the structure.
from_model_configtransformers.onnx.OnnxConfig.from_model_confighttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L130[{"name": "config", "val": ": PretrainedConfig"}, {"name": "task", "val": ": str = 'default'"}]- **config** -- The model's configuration to use when exporting to ONNX0OnnxConfig for this model Instantiate a OnnxConfig for a specific model
generate_dummy_inputstransformers.onnx.OnnxConfig.generate_dummy_inputshttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L283[{"name": "preprocessor", "val": ": typing.Union[ForwardRef('PreTrainedTokenizerBase'), ForwardRef('FeatureExtractionMixin'), ForwardRef('ImageProcessingMixin')]"}, {"name": "batch_size", "val": ": int = -1"}, {"name": "seq_length", "val": ": int = -1"}, {"name": "num_choices", "val": ": int = -1"}, {"name": "is_pair", "val": ": bool = False"}, {"name": "framework", "val": ": typing.Optional[transformers.utils.generic.TensorType] = None"}, {"name": "num_channels", "val": ": int = 3"}, {"name": "image_width", "val": ": int = 40"}, {"name": "image_height", "val": ": int = 40"}, {"name": "sampling_rate", "val": ": int = 22050"}, {"name": "time_duration", "val": ": float = 5.0"}, {"name": "frequency", "val": ": int = 220"}, {"name": "tokenizer", "val": ": typing.Optional[ForwardRef('PreTrainedTokenizerBase')] = None"}]- **preprocessor** -- ([PreTrainedTokenizerBase](/docs/transformers/v4.57.0/en/internal/tokenization_utils#transformers.PreTrainedTokenizerBase), [FeatureExtractionMixin](/docs/transformers/v4.57.0/en/main_classes/feature_extractor#transformers.FeatureExtractionMixin), or [ImageProcessingMixin](/docs/transformers/v4.57.0/en/main_classes/image_processor#transformers.ImageProcessingMixin)): The preprocessor associated with this model configuration. - **batch_size** (`int`, *optional*, defaults to -1) -- The batch size to export the model for (-1 means dynamic axis). - **num_choices** (`int`, *optional*, defaults to -1) -- The number of candidate answers provided for multiple choice task (-1 means dynamic axis). - **seq_length** (`int`, *optional*, defaults to -1) -- The sequence length to export the model for (-1 means dynamic axis). - **is_pair** (`bool`, *optional*, defaults to `False`) -- Indicate if the input is a pair (sentence 1, sentence 2) - **framework** (`TensorType`, *optional*, defaults to `None`) -- The framework (PyTorch or TensorFlow) that the tokenizer will generate tensors for. - **num_channels** (`int`, *optional*, defaults to 3) -- The number of channels of the generated images. - **image_width** (`int`, *optional*, defaults to 40) -- The width of the generated images. - **image_height** (`int`, *optional*, defaults to 40) -- The height of the generated images. - **sampling_rate** (`int`, *optional* defaults to 22050) -- The sampling rate for audio data generation. - **time_duration** (`float`, *optional* defaults to 5.0) -- Total seconds of sampling for audio data generation. - **frequency** (`int`, *optional* defaults to 220) -- The desired natural frequency of generated audio.0Mapping[str, Tensor] holding the kwargs to provide to the model's forward function Generate inputs to provide to the ONNX exporter for the specific framework
generate_dummy_inputs_onnxruntimetransformers.onnx.OnnxConfig.generate_dummy_inputs_onnxruntimehttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L403[{"name": "reference_model_inputs", "val": ": Mapping"}]- **reference_model_inputs** ([`Mapping[str, Tensor]`) -- Reference inputs for the model.0`Mapping[str, Tensor]`The mapping holding the kwargs to provide to the model's forward function Generate inputs for ONNX Runtime using the reference model inputs. Override this to run inference with seq2seq models which have the encoder and decoder exported as separate ONNX files.
use_external_data_formattransformers.onnx.OnnxConfig.use_external_data_formathttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L244[{"name": "num_parameters", "val": ": int"}]- **num_parameters** -- Number of parameter on the model0True if model.num_parameters() * size_of(float32) >= 2Gb False otherwise Flag indicating if the model requires using external data format
### OnnxConfigWithPast[[transformers.onnx.OnnxConfigWithPast]]
class transformers.onnx.OnnxConfigWithPasttransformers.onnx.OnnxConfigWithPasthttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L446[{"name": "config", "val": ": PretrainedConfig"}, {"name": "task", "val": ": str = 'default'"}, {"name": "patching_specs", "val": ": typing.Optional[list[transformers.onnx.config.PatchingSpec]] = None"}, {"name": "use_past", "val": ": bool = False"}]
fill_with_past_key_values_transformers.onnx.OnnxConfigWithPast.fill_with_past_key_values_https://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L553[{"name": "inputs_or_outputs", "val": ": Mapping"}, {"name": "direction", "val": ": str"}, {"name": "inverted_values_shape", "val": ": bool = False"}]- **inputs_or_outputs** -- The mapping to fill. - **direction** -- either "inputs" or "outputs", it specifies whether input_or_outputs is the input mapping or the output mapping, this is important for axes naming. - **inverted_values_shape** -- If `True`, store values on dynamic axis 1, else on axis 2.0 Fill the input_or_outputs mapping with past_key_values dynamic axes considering.
with_pasttransformers.onnx.OnnxConfigWithPast.with_pasthttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L457[{"name": "config", "val": ": PretrainedConfig"}, {"name": "task", "val": ": str = 'default'"}]- **config** -- The underlying model's config to use when exporting to ONNX0OnnxConfig with `.use_past = True` Instantiate a OnnxConfig with `use_past` attribute set to True
### OnnxSeq2SeqConfigWithPast[[transformers.onnx.OnnxSeq2SeqConfigWithPast]]
class transformers.onnx.OnnxSeq2SeqConfigWithPasttransformers.onnx.OnnxSeq2SeqConfigWithPasthttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/config.py#L593[{"name": "config", "val": ": PretrainedConfig"}, {"name": "task", "val": ": str = 'default'"}, {"name": "patching_specs", "val": ": typing.Optional[list[transformers.onnx.config.PatchingSpec]] = None"}, {"name": "use_past", "val": ": bool = False"}]
## ONNX Features Each ONNX configuration is associated with a set of _features_ that enable you to export models for different types of topologies or tasks. ### FeaturesManager[[transformers.onnx.FeaturesManager]]
class transformers.onnx.FeaturesManagertransformers.onnx.FeaturesManagerhttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L85[]
check_supported_model_or_raisetransformers.onnx.FeaturesManager.check_supported_model_or_raisehttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L711[{"name": "model", "val": ": typing.Union[ForwardRef('PreTrainedModel'), ForwardRef('TFPreTrainedModel')]"}, {"name": "feature", "val": ": str = 'default'"}]- **model** -- The model to export. - **feature** -- The name of the feature to check if it is available.0(str) The type of the model (OnnxConfig) The OnnxConfig instance holding the model export properties. Check whether or not the model has the requested features.
determine_frameworktransformers.onnx.FeaturesManager.determine_frameworkhttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L628[{"name": "model", "val": ": str"}, {"name": "framework", "val": ": typing.Optional[str] = None"}]- **model** (`str`) -- The name of the model to export. - **framework** (`str`, *optional*, defaults to `None`) -- The framework to use for the export. See above for priority if none provided.0The framework to use for the export. Determines the framework to use for the export. The priority is in the following order: 1. User input via `framework`. 2. If local checkpoint is provided, use the same framework as the checkpoint. 3. Available framework in environment, with priority given to PyTorch
get_configtransformers.onnx.FeaturesManager.get_confighttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L736[{"name": "model_type", "val": ": str"}, {"name": "feature", "val": ": str"}]- **model_type** (`str`) -- The model type to retrieve the config for. - **feature** (`str`) -- The feature to retrieve the config for.0`OnnxConfig`config for the combination Gets the OnnxConfig for a model_type and feature combination.
get_model_class_for_featuretransformers.onnx.FeaturesManager.get_model_class_for_featurehttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L601[{"name": "feature", "val": ": str"}, {"name": "framework", "val": ": str = 'pt'"}]- **feature** (`str`) -- The feature required. - **framework** (`str`, *optional*, defaults to `"pt"`) -- The framework to use for the export.0The AutoModel class corresponding to the feature. Attempts to retrieve an AutoModel class from a feature name.
get_model_from_featuretransformers.onnx.FeaturesManager.get_model_from_featurehttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L678[{"name": "feature", "val": ": str"}, {"name": "model", "val": ": str"}, {"name": "framework", "val": ": typing.Optional[str] = None"}, {"name": "cache_dir", "val": ": typing.Optional[str] = None"}]- **feature** (`str`) -- The feature required. - **model** (`str`) -- The name of the model to export. - **framework** (`str`, *optional*, defaults to `None`) -- The framework to use for the export. See `FeaturesManager.determine_framework` for the priority should none be provided.0The instance of the model. Attempts to retrieve a model from a model's name and the feature to be enabled.
get_supported_features_for_model_typetransformers.onnx.FeaturesManager.get_supported_features_for_model_typehttps://github.com/huggingface/transformers/blob/v4.57.0/src/transformers/onnx/features.py#L556[{"name": "model_type", "val": ": str"}, {"name": "model_name", "val": ": typing.Optional[str] = None"}]- **model_type** (`str`) -- The model type to retrieve the supported features for. - **model_name** (`str`, *optional*) -- The name attribute of the model object, only used for the exception message.0The dictionary mapping each feature to a corresponding OnnxConfig constructor. Tries to retrieve the feature -> OnnxConfig constructor map from the model type.