Training with onnxruntime and pytorch¶

ORTModule

ORTModule ¶

class onnxruntime.training.ortmodule.ORTModule(module, debug_options=None)¶

Extends user’s torch.nn.Module model to leverage ONNX Runtime super fast training engine.

ORTModule specializes the user’s torch.nn.Module model, providing forward(), backward() along with all others torch.nn.Module’s APIs.

Parameters

module (torch.nn.Module) – User’s PyTorch module that ORTModule specializes
debug_options (DebugOptions, optional) – debugging options for ORTModule.

Initializes internal Module state, shared by both nn.Module and ScriptModule.

__annotations__ = {'__call__': typing.Callable[..., typing.Any], '_is_full_backward_hook': typing.Optional[bool], '_version': <class 'int'>, 'dump_patches': <class 'bool'>, 'forward': typing.Callable[..., typing.Any], 'training': <class 'bool'>}¶

__call__(*input, **kwargs)¶: Call self as a function.

__class__¶: alias of type

__delattr__(name)¶: Implement delattr(self, name).

__dir__()¶: Default dir() implementation.

__eq__(value, /)¶: Return self==value.

__format__(format_spec, /)¶: Default object formatter.

__ge__(value, /)¶: Return self>=value.

__getattr__(name: str)¶

__getattribute__(name, /)¶: Return getattr(self, name).

__getstate__()¶

__gt__(value, /)¶: Return self>value.

__hash__()¶: Return hash(self).

__init__(module, debug_options=None)¶: Initializes internal Module state, shared by both nn.Module and ScriptModule.

__init_subclass__()¶

This method is called when a class is subclassed.

The default implementation does nothing. It may be overridden to extend subclasses.

__le__(value, /)¶: Return self<=value.

__lt__(value, /)¶: Return self<value.

__ne__(value, /)¶: Return self!=value.

__new__(**kwargs)¶

__reduce__()¶: Helper for pickle.

__reduce_ex__(protocol, /)¶: Helper for pickle.

__repr__()¶: Return repr(self).

__setattr__(name: str, value) → None¶: Implement setattr(self, name, value).

__setstate__(state)¶

__sizeof__()¶: Size of object in memory, in bytes.

__str__()¶: Return str(self).

__subclasshook__()¶

Abstract classes can override this to customize issubclass().

This is invoked early on by abc.ABCMeta.__subclasscheck__(). It should return True, False or NotImplemented. If it returns NotImplemented, the normal algorithm is used. Otherwise, it overrides the normal algorithm (and the outcome is cached).

_apply(fn)¶: Override original method to delegate execution to the flattened PyTorch user module

_call_impl(*input, **kwargs)¶

_get_backward_hooks()¶: Returns the backward hooks for use in the call function. It returns two lists, one with the full backward hooks and one with the non-full backward hooks.

_get_name()¶

_is_full_backward_hook: Optional[bool]¶

_is_training()¶

_load_from_state_dict(state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs)¶: Override original method to delegate execution to the original PyTorch user module

_maybe_warn_non_full_backward_hook(inputs, result, grad_fn)¶

_named_members(get_members_fn, prefix='', recurse=True)¶: Helper method for yielding various names + members of modules.

_register_load_state_dict_pre_hook(hook, with_module=False)¶

These hooks will be called with arguments: state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs, before loading state_dict into self. These arguments are exactly the same as those of _load_from_state_dict.

If with_module is True, then the first argument to the hook is an instance of the module.

Parameters

hook (Callable) – Callable hook that will be invoked before loading the state dict.
with_module (bool, optional) – Whether or not to pass the module instance to the hook as the first parameter.

_register_state_dict_hook(hook)¶: These hooks will be called with arguments: self, state_dict, prefix, local_metadata, after the state_dict of self is set. Note that only parameters and buffers of self or its children are guaranteed to exist in state_dict. The hooks may modify state_dict inplace or return a new one.

_replicate_for_data_parallel()¶

Raises a NotImplementedError exception since ORTModule is not compatible with torch.nn.DataParallel

torch.nn.DataParallel requires the model to be replicated across multiple devices, and in this process, ORTModule tries to export the model to onnx on multiple devices with the same sample input. Because of this multiple device export with the same sample input, torch throws an exception that reads: “RuntimeError: Input, output and indices must be on the current device” which can be vague to the user since they might not be aware of what happens behind the scene.

We therefore try to preemptively catch use of ORTModule with torch.nn.DataParallel and throw a more meaningful exception.

Users must use torch.nn.parallel.DistributedDataParallel instead of torch.nn.DataParallel which does not need model replication and is also recommended by torch to use instead.

_save_to_state_dict(destination, prefix, keep_vars)¶

Saves module state to destination dictionary, containing a state of the module, but not its descendants. This is called on every submodule in state_dict().

In rare cases, subclasses can achieve class-specific behavior by overriding this method with custom logic.

Parameters

destination (dict) – a dict where state will be stored
prefix (str) – the prefix for parameters and buffers used in this module

_slow_forward(*input, **kwargs)¶

_version: int = 1¶

add_module(name: str, module: Optional[Module]) → None¶: Raises a ORTModuleTorchModelException exception since ORTModule does not support adding modules to it

apply(fn: Callable[[Module], None]) → onnxruntime.training.ortmodule.ortmodule.T¶: Override apply() to delegate execution to ONNX Runtime

bfloat16() → torch.nn.modules.module.T¶

Casts all floating point parameters and buffers to bfloat16 datatype.

Note

This method modifies the module in-place.

Returns: self
Return type: Module

buffers(recurse: bool = True) → Iterator[torch.Tensor]¶: Override buffers()

children() → Iterator[torch.nn.modules.module.Module]¶

Returns an iterator over immediate children modules.

Yields: Module – a child module

cpu() → torch.nn.modules.module.T¶

Moves all model parameters and buffers to the CPU.

Note

This method modifies the module in-place.

Returns: self
Return type: Module

cuda(device: Optional[Union[int, torch.device]] = None) → torch.nn.modules.module.T¶

Moves all model parameters and buffers to the GPU.

This also makes associated parameters and buffers different objects. So it should be called before constructing optimizer if the module will live on GPU while being optimized.

Note

This method modifies the module in-place.

Parameters: device (int, optional) – if specified, all parameters will be copied to that device
Returns: self
Return type: Module

double() → torch.nn.modules.module.T¶

Casts all floating point parameters and buffers to double datatype.

Note

This method modifies the module in-place.

Returns: self
Return type: Module

eval() → torch.nn.modules.module.T¶

Sets the module in evaluation mode.

This has any effect only on certain modules. See documentations of particular modules for details of their behaviors in training/evaluation mode, if they are affected, e.g. Dropout, BatchNorm, etc.

This is equivalent with self.train(False).

See Locally disabling gradient computation for a comparison between .eval() and several similar mechanisms that may be confused with it.

Returns: self
Return type: Module

extra_repr() → str¶

Set the extra representation of the module

To print customized extra information, you should re-implement this method in your own modules. Both single-line and multi-line strings are acceptable.

float() → torch.nn.modules.module.T¶

Casts all floating point parameters and buffers to float datatype.

Note

This method modifies the module in-place.

Returns: self
Return type: Module

forward(*inputs, **kwargs)¶

Delegate the forward() pass of PyTorch training to ONNX Runtime.

The first call to forward performs setup and checking steps. During this call, ORTModule determines whether the module can be trained with ONNX Runtime. For this reason, the first forward call execution takes longer than subsequent calls. Execution is interupted if ONNX Runtime cannot process the model for training.

Parameters

positional (variable) –
positional –
keyword –
forward (and variable keyword arguments defined in the user's PyTorch module's) –
types. (method. Values can be torch tensors and primitive) –

Returns

The output as expected from the forward method defined by the user’s PyTorch module. Output values supported include tensors, nested sequences of tensors and nested dictionaries of tensor values.

get_buffer(target: str) → torch.Tensor¶: Override get_buffer()

get_extra_state() → Any¶

Returns any extra state to include in the module’s state_dict. Implement this and a corresponding set_extra_state() for your module if you need to store extra state. This function is called when building the module’s state_dict().

Note that extra state should be pickleable to ensure working serialization of the state_dict. We only provide provide backwards compatibility guarantees for serializing Tensors; other objects may break backwards compatibility if their serialized pickled form changes.

Returns: Any extra state to store in the module’s state_dict
Return type: object

get_parameter(target: str) → torch.nn.parameter.Parameter¶: Override get_parameter()

get_submodule(target: str) → torch.nn.modules.module.Module¶

Returns the submodule given by target if it exists, otherwise throws an error.

For example, let’s say you have an nn.Module A that looks like this:

(The diagram shows an nn.Module A. A has a nested submodule net_b, which itself has two submodules net_c and linear. net_c then has a submodule conv.)

To check whether or not we have the linear submodule, we would call get_submodule("net_b.linear"). To check whether we have the conv submodule, we would call get_submodule("net_b.net_c.conv").

The runtime of get_submodule is bounded by the degree of module nesting in target. A query against named_modules achieves the same result, but it is O(N) in the number of transitive modules. So, for a simple check to see if some submodule exists, get_submodule should always be used.

Parameters: target – The fully-qualified string name of the submodule to look for. (See above example for how to specify a fully-qualified string.)
Returns: The submodule referenced by target
Return type: torch.nn.Module
Raises: AttributeError – If the target string references an invalid path or resolves to something that is not an nn.Module

half() → torch.nn.modules.module.T¶

Casts all floating point parameters and buffers to half datatype.

Note

This method modifies the module in-place.

Returns: self
Return type: Module

load_state_dict(state_dict: OrderedDict[str, Tensor], strict: bool = True)¶: Override load_state_dict() to delegate execution to ONNX Runtime

property module¶

The original torch.nn.Module that this module wraps.

This property provides access to methods and properties on the original module.

modules() → Iterator[Module]¶: Override modules()

named_buffers(prefix: str = '', recurse: bool = True) → Iterator[Tuple[str, torch.Tensor]]¶: Override named_buffers()

named_children() → Iterator[Tuple[str, Module]]¶: Override named_children()

named_modules(*args, **kwargs)¶: Override named_modules()

named_parameters(prefix: str = '', recurse: bool = True) → Iterator[Tuple[str, torch.nn.parameter.Parameter]]¶: Override named_parameters()

parameters(recurse: bool = True) → Iterator[torch.nn.parameter.Parameter]¶: Override parameters()

register_backward_hook(hook: Callable[[torch.nn.modules.module.Module, Union[Tuple[torch.Tensor, ...], torch.Tensor], Union[Tuple[torch.Tensor, ...], torch.Tensor]], Union[None, torch.Tensor]]) → torch.utils.hooks.RemovableHandle¶

Registers a backward hook on the module.

This function is deprecated in favor of register_full_backward_hook() and the behavior of this function will change in future versions.

Returns: a handle that can be used to remove the added hook by calling handle.remove()
Return type: torch.utils.hooks.RemovableHandle

register_buffer(name: str, tensor: Optional[torch.Tensor], persistent: bool = True) → None¶: Override register_buffer()

register_forward_hook(hook: Callable[[...], None]) → torch.utils.hooks.RemovableHandle¶

Registers a forward hook on the module.

The hook will be called every time after forward() has computed an output. It should have the following signature:

hook(module, input, output) -> None or modified output

The input contains only the positional arguments given to the module. Keyword arguments won’t be passed to the hooks and only to the forward. The hook can modify the output. It can modify the input inplace but it will not have effect on forward since this is called after forward() is called.

Returns: a handle that can be used to remove the added hook by calling handle.remove()
Return type: torch.utils.hooks.RemovableHandle

register_forward_pre_hook(hook: Callable[[...], None]) → torch.utils.hooks.RemovableHandle¶

Registers a forward pre-hook on the module.

The hook will be called every time before forward() is invoked. It should have the following signature:

hook(module, input) -> None or modified input

The input contains only the positional arguments given to the module. Keyword arguments won’t be passed to the hooks and only to the forward. The hook can modify the input. User can either return a tuple or a single modified value in the hook. We will wrap the value into a tuple if a single value is returned(unless that value is already a tuple).

Returns: a handle that can be used to remove the added hook by calling handle.remove()
Return type: torch.utils.hooks.RemovableHandle

register_full_backward_hook(hook: Callable[[torch.nn.modules.module.Module, Union[Tuple[torch.Tensor, ...], torch.Tensor], Union[Tuple[torch.Tensor, ...], torch.Tensor]], Union[None, torch.Tensor]]) → torch.utils.hooks.RemovableHandle¶

Registers a backward hook on the module.

The hook will be called every time the gradients with respect to module inputs are computed. The hook should have the following signature:

hook(module, grad_input, grad_output) -> tuple(Tensor) or None

The grad_input and grad_output are tuples that contain the gradients with respect to the inputs and outputs respectively. The hook should not modify its arguments, but it can optionally return a new gradient with respect to the input that will be used in place of grad_input in subsequent computations. grad_input will only correspond to the inputs given as positional arguments and all kwarg arguments are ignored. Entries in grad_input and grad_output will be None for all non-Tensor arguments.

For technical reasons, when this hook is applied to a Module, its forward function will receive a view of each Tensor passed to the Module. Similarly the caller will receive a view of each Tensor returned by the Module’s forward function.

Warning

Modifying inputs or outputs inplace is not allowed when using backward hooks and will raise an error.

Returns: a handle that can be used to remove the added hook by calling handle.remove()
Return type: torch.utils.hooks.RemovableHandle

register_parameter(name: str, param: Optional[torch.nn.parameter.Parameter]) → None¶: Override register_parameter()

requires_grad_(requires_grad: bool = True) → torch.nn.modules.module.T¶

Change if autograd should record operations on parameters in this module.

This method sets the parameters’ requires_grad attributes in-place.

This method is helpful for freezing part of the module for finetuning or training parts of a model individually (e.g., GAN training).

See Locally disabling gradient computation for a comparison between .requires_grad_() and several similar mechanisms that may be confused with it.

Parameters: requires_grad (bool) – whether autograd should record operations on parameters in this module. Default: True.
Returns: self
Return type: Module

set_extra_state(state: Any)¶

This function is called from load_state_dict() to handle any extra state found within the state_dict. Implement this function and a corresponding get_extra_state() for your module if you need to store extra state within its state_dict.

Parameters: state (dict) – Extra state from the state_dict

share_memory() → torch.nn.modules.module.T¶: See torch.Tensor.share_memory_()

state_dict(destination=None, prefix='', keep_vars=False)¶: Override state_dict() to delegate execution to ONNX Runtime

to(*args, **kwargs)¶

Moves and/or casts the parameters and buffers.

This can be called as

to(device=None, dtype=None, non_blocking=False)

to(dtype, non_blocking=False)

to(tensor, non_blocking=False)

to(memory_format=torch.channels_last)

Its signature is similar to torch.Tensor.to(), but only accepts floating point or complex dtypes. In addition, this method will only cast the floating point or complex parameters and buffers to dtype (if given). The integral parameters and buffers will be moved device, if that is given, but with dtypes unchanged. When non_blocking is set, it tries to convert/move asynchronously with respect to the host if possible, e.g., moving CPU Tensors with pinned memory to CUDA devices.

See below for examples.

Note

This method modifies the module in-place.

Parameters

device (torch.device) – the desired device of the parameters and buffers in this module
dtype (torch.dtype) – the desired floating point or complex dtype of the parameters and buffers in this module
tensor (torch.Tensor) – Tensor whose dtype and device are the desired dtype and device for all parameters and buffers in this module
memory_format (torch.memory_format) – the desired memory format for 4D parameters and buffers in this module (keyword only argument)

Returns

self

Return type

Module

Examples:

>>> linear = nn.Linear(2, 2)
>>> linear.weight
Parameter containing:
tensor([[ 0.1913, -0.3420],
        [-0.5113, -0.2325]])
>>> linear.to(torch.double)
Linear(in_features=2, out_features=2, bias=True)
>>> linear.weight
Parameter containing:
tensor([[ 0.1913, -0.3420],
        [-0.5113, -0.2325]], dtype=torch.float64)
>>> gpu1 = torch.device("cuda:1")
>>> linear.to(gpu1, dtype=torch.half, non_blocking=True)
Linear(in_features=2, out_features=2, bias=True)
>>> linear.weight
Parameter containing:
tensor([[ 0.1914, -0.3420],
        [-0.5112, -0.2324]], dtype=torch.float16, device='cuda:1')
>>> cpu = torch.device("cpu")
>>> linear.to(cpu)
Linear(in_features=2, out_features=2, bias=True)
>>> linear.weight
Parameter containing:
tensor([[ 0.1914, -0.3420],
        [-0.5112, -0.2324]], dtype=torch.float16)

>>> linear = nn.Linear(2, 2, bias=None).to(torch.cdouble)
>>> linear.weight
Parameter containing:
tensor([[ 0.3741+0.j,  0.2382+0.j],
        [ 0.5593+0.j, -0.4443+0.j]], dtype=torch.complex128)
>>> linear(torch.ones(3, 2, dtype=torch.cdouble))
tensor([[0.6122+0.j, 0.1150+0.j],
        [0.6122+0.j, 0.1150+0.j],
        [0.6122+0.j, 0.1150+0.j]], dtype=torch.complex128)

to_empty(*, device: Union[str, torch.device]) → torch.nn.modules.module.T¶

Moves the parameters and buffers to the specified device without copying storage.

Parameters: device (torch.device) – The desired device of the parameters and buffers in this module.
Returns: self
Return type: Module

train(mode: bool = True) → onnxruntime.training.ortmodule.ortmodule.T¶: Override train() to delegate execution to ONNX Runtime

type(dst_type: Union[torch.dtype, str]) → torch.nn.modules.module.T¶

Casts all parameters and buffers to dst_type.

Note

This method modifies the module in-place.

Parameters: dst_type (type or string) – the desired type
Returns: self
Return type: Module

xpu(device: Optional[Union[int, torch.device]] = None) → torch.nn.modules.module.T¶

Moves all model parameters and buffers to the XPU.

This also makes associated parameters and buffers different objects. So it should be called before constructing optimizer if the module will live on XPU while being optimized.

Note

This method modifies the module in-place.

Parameters: device (int, optional) – if specified, all parameters will be copied to that device
Returns: self
Return type: Module

zero_grad(set_to_none: bool = False) → None¶

Sets gradients of all model parameters to zero. See similar function under torch.optim.Optimizer for more context.

Parameters: set_to_none (bool) – instead of setting to zero, set the grads to None. See torch.optim.Optimizer.zero_grad() for details.

Partial Training

Exceptions

Training with onnxruntime and pytorch¶

ORTModule¶

ORTModule ¶