Optimum documentation

Export a model to ONNX with optimum.exporters.onnx

Hugging Face's logo
Join the Hugging Face community

and get access to the augmented documentation experience

to get started

Export a model to ONNX with optimum.exporters.onnx

Summary

Exporting a model to ONNX is as simple as

optimum-cli export onnx --model gpt2 gpt2_onnx/

Check out the help for more options:

optimum-cli export onnx --help

Why use ONNX?

If you need to deploy 🤗 Transformers or 🤗 Diffusers models in production environments, we recommend exporting them to a serialized format that can be loaded and executed on specialized runtimes and hardware. In this guide, we’ll show you how to export these models to ONNX (Open Neural Network eXchange).

ONNX is an open standard that defines a common set of operators and a common file format to represent deep learning models in a wide variety of frameworks, including PyTorch and TensorFlow. When a model is exported to the ONNX format, these operators are used to construct a computational graph (often called an intermediate representation) which represents the flow of data through the neural network.

By exposing a graph with standardized operators and data types, ONNX makes it easy to switch between frameworks. For example, a model trained in PyTorch can be exported to ONNX format and then imported in TensorRT or OpenVino.

Once exported, a model can be optimized for inference via techniques such as graph optimization and quantization. Check the optimum.onnxruntime subpackage to optimize and run ONNX models!

🤗 Optimum provides support for the ONNX export by leveraging configuration objects. These configuration objects come ready made for a number of model architectures, and are designed to be easily extendable to other architectures.

To check the supported architectures, go to the configuration reference page.

Exporting a model to ONNX using the CLI

To export a 🤗 Transformers or 🤗 Diffusers model to ONNX, you’ll first need to install some extra dependencies:

pip install optimum[exporters]

The Optimum ONNX export can be used through Optimum command-line:

optimum-cli export onnx --help

usage: optimum-cli <command> [<args>] export onnx [-h] -m MODEL [--task TASK] [--for-ort] [--device DEVICE] [--opset OPSET] [--atol ATOL]
                                                  [--framework {pt,tf}] [--pad_token_id PAD_TOKEN_ID] [--cache_dir CACHE_DIR] [--batch_size BATCH_SIZE]
                                                  [--sequence_length SEQUENCE_LENGTH] [--num_choices NUM_CHOICES] [--width WIDTH] [--height HEIGHT]
                                                  [--num_channels NUM_CHANNELS] [--feature_size FEATURE_SIZE] [--nb_max_frames NB_MAX_FRAMES]
                                                  [--audio_sequence_length AUDIO_SEQUENCE_LENGTH]
                                                  output

Required arguments:
  -m MODEL, --model MODEL
                        Model ID on huggingface.co or path on disk to load model from.
  output                Path indicating the directory where to store generated ONNX model.

Optional arguments:
  --task TASK           The task to export the model for. If not specified, the task will be auto-inferred based on the model. Available tasks depend on
                        the model, but are among: ['default', 'masked-lm', 'causal-lm', 'seq2seq-lm', 'sequence-classification', 'token-classification',
                        'multiple-choice', 'object-detection', 'question-answering', 'image-classification', 'image-segmentation', 'masked-im', 'semantic-
                        segmentation', 'speech2seq-lm', 'stable-diffusion']. For decoder models, use `xxx-with-past` to export the model using past key
                        values in the decoder.
  --for-ort             This exports models ready to be run with Optimum's ORTModel. Useful for encoder-decoder models forconditional generation. If
                        enabled the encoder and decoder of the model are exported separately.
  --device DEVICE       The device to use to do the export. Defaults to "cpu".
  --opset OPSET         If specified, ONNX opset version to export the model with. Otherwise, the default opset will be used.
  --atol ATOL           If specified, the absolute difference tolerance when validating the model. Otherwise, the default atol for the model will be used.
  --framework {pt,tf}   The framework to use for the ONNX export. If not provided, will attempt to use the local checkpoint's original framework or what is
                        available in the environment.
  --pad_token_id PAD_TOKEN_ID
                        This is needed by some models, for some tasks. If not provided, will attempt to use the tokenizer to guess it.
  --cache_dir CACHE_DIR
                        Path indicating where to store cache.
  --trust-remote-code   Allow to use custom code for the modeling hosted in the model repository. This option should only be set for repositories you trust
                        and in which you have read the code, as it will execute on your local machine arbitrary code present in the model repository.

Exporting a checkpoint can be done as follows:

optimum-cli export onnx --model distilbert-base-uncased-distilled-squad distilbert_base_uncased_squad_onnx/

You should see the following logs (along with potential logs from PyTorch / TensorFlow that were hidden here for clarity):

Automatic task detection to question-answering.
Framework not specified. Using pt to export to ONNX.
Using framework PyTorch: 1.12.1

Validating ONNX model...
        -[✓] ONNX model output names match reference model (start_logits, end_logits)
        - Validating ONNX Model output "start_logits":
                -[✓] (2, 16) matches (2, 16)
                -[✓] all values close (atol: 0.0001)
        - Validating ONNX Model output "end_logits":
                -[✓] (2, 16) matches (2, 16)
                -[✓] all values close (atol: 0.0001)
All good, model saved at: distilbert_base_uncased_squad_onnx/model.onnx

This exports an ONNX graph of the checkpoint defined by the --model argument. As you can see, the task was automatically detected. This was possible because the model was on the Hub.

For local models, providing the --task argument is needed or it will default to the model architecture without any task specific head:

optimum-cli export onnx --model local_path --task question-answering distilbert_base_uncased_squad_onnx/

Note that providing the --task argument for a model on the Hub will disable the automatic task detection.

The resulting model.onnx file can then be run on one of the many accelerators that support the ONNX standard. For example, we can load and run the model with ONNX Runtime using the optimum.onnxruntime package as follows:

>>> from transformers import AutoTokenizer
>>> from optimum.onnxruntime import ORTModelForQuestionAnswering

>>> tokenizer = AutoTokenizer.from_pretrained("distilbert_base_uncased_squad_onnx")
>>> model = ORTModelForQuestionAnswering.from_pretrained("distilbert_base_uncased_squad_onnx")
>>> inputs = tokenizer("What am I using?", "Using DistilBERT with ONNX Runtime!", return_tensors="pt")
>>> outputs = model(**inputs)

Printing the outputs would give that:

QuestionAnsweringModelOutput(loss=None, start_logits=tensor([[-4.7652, -1.0452, -7.0409, -4.6864, -4.0277, -6.2021, -4.9473,  2.6287,
          7.6111, -1.2488, -2.0551, -0.9350,  4.9758, -0.7707,  2.1493, -2.0703,
         -4.3232, -4.9472]]), end_logits=tensor([[ 0.4382, -1.6502, -6.3654, -6.0661, -4.1482, -3.5779, -0.0774, -3.6168,
         -1.8750, -2.8910,  6.2582,  0.5425, -3.7699,  3.8232, -1.5073,  6.2311,
          3.3604, -0.0772]]), hidden_states=None, attentions=None)

As you can see, converting a model to ONNX does not mean leaving the Hugging Face ecosystem. You end up with a similar API as regular 🤗 Transformers models!

It is also possible to export the model to ONNX directly from the ORTModelForQuestionAnswering class by doing the following:

>>> model = ORTModelForQuestionAnswering.from_pretrained("distilbert-base-uncased-distilled-squad", from_transformers=True)

For more information, check the optimum.onnxruntime documentation page on this topic.

The process is identical for TensorFlow checkpoints on the Hub. For example, we can export a pure TensorFlow checkpoint from the Keras organization as follows:

optimum-cli export onnx --model keras-io/transformers-qa distilbert_base_cased_squad_onnx/

Exporting a model to be used with Optimum's ORTModel

Models exported through optimum-cli export onnx can be used directly in ORTModel by passing the parameter --for-ort. This is especially useful for encoder-decoder models, where in this case the export will split the encoder and decoder into two .onnx files, as the encoder is usually only run once while the decoder may be run several times in autogenerative tasks.

Exporting a model using past keys/values in the decoder

When exporting a decoder model used for generation, it can be useful to encapsulate in the exported ONNX the reuse of past keys and values. This allows to avoid recomputing the same intermediate activations during the generation.

In the ONNX export, the past keys/values are reused by default. This behavior corresponds to --task seq2seq-lm-with-past, --task causal-lm-with-past, or --task speech2seq-lm-with-past. If for any purpose you would like to disable the export with past keys/values reuse, passing explicitly to optimum-cli export onnx the task seq2seq-lm, causal-lm or speech2seq-lm is required.

A model exported using past key/values can be reused directly into Optimum’s ORTModel:

optimum-cli export onnx --model gpt2 --for-ort --task causal-lm-with-past gpt2_onnx/

and

>>> from transformers import AutoTokenizer
>>> from optimum.onnxruntime import ORTModelForCausalLM

>>> tokenizer = AutoTokenizer.from_pretrained("./gpt2_onnx/")
>>> model = ORTModelForCausalLM.from_pretrained("./gpt2_onnx/")
>>> inputs = tokenizer("My name is Arthur and I live in", return_tensors="pt")
>>> gen_tokens = model.generate(**inputs)
>>> print(tokenizer.batch_decode(gen_tokens))
# prints ['My name is Arthur and I live in the United States of America. I am a member of the']

Selecting a task

Specifying a --task should not be necessary in most cases when exporting from a model on the Hugging Face Hub.

However, in case you need to check for a given a model architecture what tasks the ONNX export supports, we got you covered. First, you can check the list of supported tasks for both PyTorch and TensorFlow here.

For each model architecture, you can find the list of supported tasks via the TasksManager. For example, for DistilBERT, for the ONNX export, we have:

>>> from optimum.exporters.tasks import TasksManager

>>> distilbert_tasks = list(TasksManager.get_supported_tasks_for_model_type("distilbert", "onnx").keys())
>>> print(distilbert_tasks)
['default', 'masked-lm', 'sequence-classification', 'multiple-choice', 'token-classification', 'question-answering']

You can then pass one of these tasks to the --task argument in the optimum-cli export onnx command, as mentioned above.