API¶

Data on CPU¶

On CPU (the default), OrtValues can be mapped to and from native Python data structures: numpy arrays, dictionaries and lists of numpy arrays.

# X is numpy array on cpu
ortvalue = onnxruntime.OrtValue.ortvalue_from_numpy(X)
ortvalue.device_name()  # 'cpu'
ortvalue.shape()        # shape of the numpy array X
ortvalue.data_type()    # 'tensor(float)'
ortvalue.is_tensor()    # 'True'
np.array_equal(ortvalue.numpy(), X)  # 'True'

# ortvalue can be provided as part of the input feed to a model
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
results = session.run(["Y"], {"X": ortvalue})

By default, ONNX Runtime always places input(s) and output(s) on CPU. Having the data on CPU may not optimal if the input or output is consumed and produced on a device other than CPU because it introduces data copy between CPU and the device.

Data on device¶

ONNX Runtime supports a custom data structure that supports all ONNX data formats that allows users to place the data backing these on a device, for example, on a CUDA supported device. In ONNX Runtime, this called IOBinding.

To use the IOBinding feature, replace InferenceSession.run() with InferenceSession.run_with_iobinding().

A graph is executed on a device other than CPU, for instance CUDA. Users can use IOBinding to copy the data onto the GPU.

# X is numpy array on cpu
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
io_binding = session.io_binding()
# OnnxRuntime will copy the data over to the CUDA device if 'input' is consumed by nodes on the CUDA device
io_binding.bind_cpu_input('input', X)
io_binding.bind_output('output')
session.run_with_iobinding(io_binding)
Y = io_binding.copy_outputs_to_cpu()[0]

The input data is on a device, users directly use the input. The output data is on CPU.

# X is numpy array on cpu
X_ortvalue = onnxruntime.OrtValue.ortvalue_from_numpy(X, 'cuda', 0)
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
io_binding = session.io_binding()
io_binding.bind_input(name='input', device_type=X_ortvalue.device_name(), device_id=0, element_type=np.float32, shape=X_ortvalue.shape(), buffer_ptr=X_ortvalue.data_ptr())
io_binding.bind_output('output')
session.run_with_iobinding(io_binding)
Y = io_binding.copy_outputs_to_cpu()[0]

The input data and output data are both on a device, users directly use the input and also place output on the device.

#X is numpy array on cpu
X_ortvalue = onnxruntime.OrtValue.ortvalue_from_numpy(X, 'cuda', 0)
Y_ortvalue = onnxruntime.OrtValue.ortvalue_from_shape_and_type([3, 2], np.float32, 'cuda', 0)  # Change the shape to the actual shape of the output being bound
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
io_binding = session.io_binding()
io_binding.bind_input(name='input', device_type=X_ortvalue.device_name(), device_id=0, element_type=np.float32, shape=X_ortvalue.shape(), buffer_ptr=X_ortvalue.data_ptr())
io_binding.bind_output(name='output', device_type=Y_ortvalue.device_name(), device_id=0, element_type=np.float32, shape=Y_ortvalue.shape(), buffer_ptr=Y_ortvalue.data_ptr())
session.run_with_iobinding(io_binding)

Users can request ONNX Runtime to allocate an output on a device. This is particularly useful for dynamic shaped outputs. Users can use the get_outputs() API to get access to the OrtValue (s) corresponding to the allocated output(s). Users can thus consume the ONNX Runtime allocated memory for the output as an OrtValue.

#X is numpy array on cpu
X_ortvalue = onnxruntime.OrtValue.ortvalue_from_numpy(X, 'cuda', 0)
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
io_binding = session.io_binding()
io_binding.bind_input(name='input', device_type=X_ortvalue.device_name(), device_id=0, element_type=np.float32, shape=X_ortvalue.shape(), buffer_ptr=X_ortvalue.data_ptr())
#Request ONNX Runtime to bind and allocate memory on CUDA for 'output'
io_binding.bind_output('output', 'cuda')
session.run_with_iobinding(io_binding)
# The following call returns an OrtValue which has data allocated by ONNX Runtime on CUDA
ort_output = io_binding.get_outputs()[0]

In addition, ONNX Runtime supports directly working with OrtValue (s) while inferencing a model if provided as part of the input feed.

Users can bind OrtValue (s) directly.

#X is numpy array on cpu
#X is numpy array on cpu
X_ortvalue = onnxruntime.OrtValue.ortvalue_from_numpy(X, 'cuda', 0)
Y_ortvalue = onnxruntime.OrtValue.ortvalue_from_shape_and_type([3, 2], np.float32, 'cuda', 0)  # Change the shape to the actual shape of the output being bound
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
io_binding = session.io_binding()
io_binding.bind_ortvalue_input('input', X_ortvalue)
io_binding.bind_ortvalue_output('output', Y_ortvalue)
session.run_with_iobinding(io_binding)

You can also bind inputs and outputs directly to a PyTorch tensor.

# X is a PyTorch tensor on device
session = onnxruntime.InferenceSession('model.onnx', providers=['CUDAExecutionProvider', 'CPUExecutionProvider']))
binding = session.io_binding()

X_tensor = X.contiguous()

binding.bind_input(
    name='X',
    device_type='cuda',
    device_id=0,
    element_type=np.float32,
    shape=tuple(x_tensor.shape),
    buffer_ptr=x_tensor.data_ptr(),
    )

## Allocate the PyTorch tensor for the model output
Y_shape = ... # You need to specify the output PyTorch tensor shape
Y_tensor = torch.empty(Y_shape, dtype=torch.float32, device='cuda:0').contiguous()
binding.bind_output(
    name='Y',
    device_type='cuda',
    device_id=0,
    element_type=np.float32,
    shape=tuple(Y_tensor.shape),
    buffer_ptr=Y_tensor.data_ptr(),
)

session.run_with_iobinding(binding)

RunOptions¶

class onnxruntime.RunOptions(self: onnxruntime.capi.onnxruntime_pybind11_state.RunOptions) → None¶

Configuration information for a single Run.

add_run_config_entry(self: onnxruntime.capi.onnxruntime_pybind11_state.RunOptions, arg0: str, arg1: str) → None¶

Set a single run configuration entry as a pair of strings.

get_run_config_entry(self: onnxruntime.capi.onnxruntime_pybind11_state.RunOptions, arg0: str) → str¶

Get a single run configuration value using the given configuration key.

property log_severity_level¶

Log severity level for a particular Run() invocation. 0:Verbose, 1:Info, 2:Warning. 3:Error, 4:Fatal. Default is 2.

property log_verbosity_level¶

VLOG level if DEBUG build and run_log_severity_level is 0. Applies to a particular Run() invocation. Default is 0.

property logid¶

To identify logs generated by a particular Run() invocation.

property only_execute_path_to_fetches¶

Only execute the nodes needed by fetch list

property terminate¶

Set to True to terminate any currently executing calls that are using this RunOptions instance. The individual calls will exit gracefully and return an error status.

SessionOptions¶

class onnxruntime.SessionOptions(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions) → None¶

Configuration information for a session.

add_free_dimension_override_by_denotation(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str, arg1: int) → None¶

Specify the dimension size for each denotation associated with an input’s free dimension.

add_free_dimension_override_by_name(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str, arg1: int) → None¶

Specify values of named dimensions within model inputs.

add_initializer(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str, arg1: object) → None¶

add_session_config_entry(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str, arg1: str) → None¶

Set a single session configuration entry as a pair of strings.

property enable_cpu_mem_arena¶

Enables the memory arena on CPU. Arena may pre-allocate memory for future usage. Set this option to false if you don’t want it. Default is True.

property enable_mem_pattern¶

Enable the memory pattern optimization. Default is true.

property enable_mem_reuse¶

Enable the memory reuse optimization. Default is true.

property enable_profiling¶

Enable profiling for this session. Default is false.

property execution_mode¶

Sets the execution mode. Default is sequential.

property execution_order¶

Sets the execution order. Default is basic topological order.

get_session_config_entry(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str) → str¶

Get a single session configuration value using the given configuration key.

property graph_optimization_level¶

Graph optimization level for this session.

property inter_op_num_threads¶

Sets the number of threads used to parallelize the execution of the graph (across nodes). Default is 0 to let onnxruntime choose.

property intra_op_num_threads¶

Sets the number of threads used to parallelize the execution within nodes. Default is 0 to let onnxruntime choose.

property log_severity_level¶

Log severity level. Applies to session load, initialization, etc. 0:Verbose, 1:Info, 2:Warning. 3:Error, 4:Fatal. Default is 2.

property log_verbosity_level¶

VLOG level if DEBUG build and session_log_severity_level is 0. Applies to session load, initialization, etc. Default is 0.

property logid¶

Logger id to use for session output.

property optimized_model_filepath¶

File path to serialize optimized model to. Optimized model is not serialized unless optimized_model_filepath is set. Serialized model format will default to ONNX unless: - add_session_config_entry is used to set ‘session.save_model_format’ to ‘ORT’, or - there is no ‘session.save_model_format’ config entry and optimized_model_filepath ends in ‘.ort’ (case insensitive)

property profile_file_prefix¶

The prefix of the profile file. The current time will be appended to the file name.

register_custom_ops_library(self: onnxruntime.capi.onnxruntime_pybind11_state.SessionOptions, arg0: str) → None¶

Specify the path to the shared library containing the custom op kernels required to run a model.

property use_deterministic_compute¶

Whether to use deterministic compute. Default is false.

OrtValue¶

class onnxruntime.OrtValue(ortvalue, numpy_obj=None)[source]¶

A data structure that supports all ONNX data formats (tensors and non-tensors) that allows users to place the data backing these on a device, for example, on a CUDA supported device. This class provides APIs to construct and deal with OrtValues.

as_sparse_tensor()[source]¶

The function will return SparseTensor contained in this OrtValue

data_ptr()[source]¶

Returns the address of the first element in the OrtValue’s data buffer

data_type()[source]¶

Returns the data type of the data in the OrtValue

device_name()[source]¶

Returns the name of the device where the OrtValue’s data buffer resides e.g. cpu, cuda

has_value()[source]¶

Returns True if the OrtValue corresponding to an optional type contains data, else returns False

is_sparse_tensor()[source]¶

Returns True if the OrtValue contains a SparseTensor, else returns False

is_tensor()[source]¶

Returns True if the OrtValue contains a Tensor, else returns False

is_tensor_sequence()[source]¶

Returns True if the OrtValue contains a Tensor Sequence, else returns False

numpy()[source]¶

Returns a Numpy object from the OrtValue. Valid only for OrtValues holding Tensors. Throws for OrtValues holding non-Tensors. Use accessors to gain a reference to non-Tensor objects such as SparseTensor

static ort_value_from_sparse_tensor(sparse_tensor)[source]¶

The function will construct an OrtValue instance from a valid SparseTensor The new instance of OrtValue will assume the ownership of sparse_tensor

static ortvalue_from_numpy(numpy_obj, device_type='cpu', device_id=0)[source]¶

Factory method to construct an OrtValue (which holds a Tensor) from a given Numpy object A copy of the data in the Numpy object is held by the OrtValue only if the device is NOT cpu

Parameters

• numpy_obj – The Numpy object to construct the OrtValue from

• device_type – e.g. cpu, cuda, cpu by default

• device_id – device id, e.g. 0

static ortvalue_from_shape_and_type(shape=None, element_type=None, device_type='cpu', device_id=0)[source]¶

Factory method to construct an OrtValue (which holds a Tensor) from given shape and element_type

Parameters

• shape – List of integers indicating the shape of the OrtValue

• element_type – The data type of the elements in the OrtValue (numpy type)

• device_type – e.g. cpu, cuda, cpu by default

• device_id – device id, e.g. 0

shape()[source]¶

Returns the shape of the data in the OrtValue

update_inplace(np_arr)[source]¶

Update the OrtValue in place with a new Numpy array. The numpy contents are copied over to the device memory backing the OrtValue. It can be used to update the input valuess for an InferenceSession with CUDA graph enabled or other scenarios where the OrtValue needs to be updated while the memory address can not be changed.

SparseTensor¶

class onnxruntime.SparseTensor(sparse_tensor)[source]¶

A data structure that project the C++ SparseTensor object The class provides API to work with the object. Depending on the format, the class will hold more than one buffer depending on the format

Internal constructor

as_blocksparse_view()[source]¶

The method will return coo representation of the sparse tensor which will enable querying BlockSparse indices. If the instance did not contain BlockSparse format, it would throw. You can query coo indices as:

block_sparse_indices = sparse_tensor.as_blocksparse_view().indices()

which will return a numpy array that is backed by the native memory

as_coo_view()[source]¶

The method will return coo representation of the sparse tensor which will enable querying COO indices. If the instance did not contain COO format, it would throw. You can query coo indices as:

coo_indices = sparse_tensor.as_coo_view().indices()

which will return a numpy array that is backed by the native memory.

as_csrc_view()[source]¶

The method will return CSR(C) representation of the sparse tensor which will enable querying CRS(C) indices. If the instance dit not contain CSR(C) format, it would throw. You can query indices as:

inner_ndices = sparse_tensor.as_csrc_view().inner()
outer_ndices = sparse_tensor.as_csrc_view().outer()

returning numpy arrays backed by the native memory.

data_type()[source]¶

Returns a string data type of the data in the OrtValue

dense_shape()[source]¶

Returns a numpy array(int64) containing a dense shape of a sparse tensor

device_name()[source]¶

Returns the name of the device where the SparseTensor data buffers reside e.g. cpu, cuda

format()[source]¶

Returns a OrtSparseFormat enumeration

static sparse_coo_from_numpy(dense_shape, values, coo_indices, ort_device)[source]¶

Factory method to construct a SparseTensor in COO format from given arguments

Parameters

• dense_shape – 1-D numpy array(int64) or a python list that contains a dense_shape of the sparse tensor must be on cpu memory

• values – a homogeneous, contiguous 1-D numpy array that contains non-zero elements of the tensor of a type.

• coo_indices – contiguous numpy array(int64) that contains COO indices for the tensor. coo_indices may have a 1-D shape when it contains a linear index of non-zero values and its length must be equal to that of the values. It can also be of 2-D shape, in which has it contains pairs of coordinates for each of the nnz values and its length must be exactly twice of the values length.

• ort_device –

  • describes the backing memory owned by the supplied nummpy arrays. Only CPU memory is

  suppored for non-numeric data types.

For primitive types, the method will map values and coo_indices arrays into native memory and will use them as backing storage. It will increment the reference count for numpy arrays and it will decrement it on GC. The buffers may reside in any storage either CPU or GPU. For strings and objects, it will create a copy of the arrays in CPU memory as ORT does not support those on other devices and their memory can not be mapped.

static sparse_csr_from_numpy(dense_shape, values, inner_indices, outer_indices, ort_device)[source]¶

Factory method to construct a SparseTensor in CSR format from given arguments

Parameters

• dense_shape – 1-D numpy array(int64) or a python list that contains a dense_shape of the sparse tensor (rows, cols) must be on cpu memory

• values – a contiguous, homogeneous 1-D numpy array that contains non-zero elements of the tensor of a type.

• inner_indices – contiguous 1-D numpy array(int64) that contains CSR inner indices for the tensor. Its length must be equal to that of the values.

• outer_indices – contiguous 1-D numpy array(int64) that contains CSR outer indices for the tensor. Its length must be equal to the number of rows + 1.

• ort_device –

  • describes the backing memory owned by the supplied nummpy arrays. Only CPU memory is

  suppored for non-numeric data types.

For primitive types, the method will map values and indices arrays into native memory and will use them as backing storage. It will increment the reference count and it will decrement then count when it is GCed. The buffers may reside in any storage either CPU or GPU. For strings and objects, it will create a copy of the arrays in CPU memory as ORT does not support those on other devices and their memory can not be mapped.

to_cuda(ort_device)[source]¶

Returns a copy of this instance on the specified cuda device

Parameters

ort_device – with name ‘cuda’ and valid gpu device id

The method will throw if:

• this instance contains strings

• this instance is already on GPU. Cross GPU copy is not supported

• CUDA is not present in this build

• if the specified device is not valid

values()[source]¶

The method returns a numpy array that is backed by the native memory if the data type is numeric. Otherwise, the returned numpy array that contains copies of the strings.

IOBinding¶

class onnxruntime.IOBinding(session)[source]¶

This class provides API to bind input/output to a specified device, e.g. GPU.

bind_cpu_input(name, arr_on_cpu)[source]¶

bind an input to array on CPU :param name: input name :param arr_on_cpu: input values as a python array on CPU

bind_input(name, device_type, device_id, element_type, shape, buffer_ptr)[source]¶

Parameters

• name – input name

• device_type – e.g. cpu, cuda

• device_id – device id, e.g. 0

• element_type – input element type

• shape – input shape

• buffer_ptr – memory pointer to input data

bind_ortvalue_input(name, ortvalue)[source]¶

Parameters

• name – input name

• ortvalue – OrtValue instance to bind

bind_ortvalue_output(name, ortvalue)[source]¶

Parameters

• name – output name

• ortvalue – OrtValue instance to bind

bind_output(name, device_type='cpu', device_id=0, element_type=None, shape=None, buffer_ptr=None)[source]¶

Parameters

• name – output name

• device_type – e.g. cpu, cuda, cpu by default

• device_id – device id, e.g. 0

• element_type – output element type

• shape – output shape

• buffer_ptr – memory pointer to output data

copy_outputs_to_cpu()[source]¶

Copy output contents to CPU (if on another device). No-op if already on the CPU.

get_outputs()[source]¶

Returns the output OrtValues from the Run() that preceded the call. The data buffer of the obtained OrtValues may not reside on CPU memory

OrtDevice¶

class onnxruntime.OrtDevice(c_ort_device)[source]¶

A data structure that exposes the underlying C++ OrtDevice

Internal constructor

ModelMetadata¶

class onnxruntime.ModelMetadata¶

Pre-defined and custom metadata about the model. It is usually used to identify the model used to run the prediction and facilitate the comparison.

property custom_metadata_map¶

additional metadata

property description¶

description of the model

property domain¶

ONNX domain

property graph_description¶

description of the graph hosted in the model

property graph_name¶

graph name

property producer_name¶

producer name

property version¶

version of the model

NodeArg¶

class onnxruntime.NodeArg¶

Node argument definition, for both input and output, including arg name, arg type (contains both type and shape).

property name¶

node name

property shape¶

node shape (assuming the node holds a tensor)

property type¶

node type