Modulation Filterbanks 

param fc:

Center frequencies of the input auditory channels in Hz. Shape: (n_channels,).

type fc:

Tensor

param Q:

Q-factor for modulation filters (bandwidth = mfc/Q for mfc > 10 Hz). Default: 2.0.

type Q:

param max_mfc:

Maximum modulation frequency in Hz. Can be further limited by use_upper_limit. Default: 150.0.

type max_mfc:

param lp_cutoff:

Cutoff frequency for the main lowpass filter (mfc=0) in Hz. Default: 2.5 Hz (constant across all presets, per MATLAB AMT).

type lp_cutoff:

param att_factor:

Attenuation factor applied to modulation filters with mfc > 10 Hz. Default: 1.0 (no attenuation). For jepsen2008/paulick2024, use 1/√2 ≈ 0.707.

type att_factor:

param use_upper_limit:

If True, apply dynamic upper limit based on auditory channel frequency: \(\text{umf} = \min(0.25 \cdot f_c, \text{max\_mfc})\). Default: False (fixed limit at max_mfc).

type use_upper_limit:

bool

param preset:

Configuration preset selecting published model parameters:

‘dau1997’: Original CASP model (Dau et al. 1997)
- lp_cutoff = 2.5 Hz
- att_factor = 1.0
- use_upper_limit = False (fixed 150 Hz max)
- No 150 Hz pre-filtering
‘jepsen2008’: Updated CASP model (Jepsen et al. 2008)
- lp_cutoff = 2.5 Hz (main lowpass)
- att_factor = 1/√2 (≈ 0.707)
- use_upper_limit = True (dynamic limit = 0.25 × fc)
- max_mfc = 150 Hz
- 150 Hz pre-filtering enabled
‘paulick2024’: Revised CASP model (Paulick et al. 2024)
- Same configuration as ‘jepsen2008’
- Derived from paulick2024 model for reusability

If preset is provided, it overrides lp_cutoff, att_factor, use_upper_limit, and max_mfc (unless explicitly provided as arguments). Default: None (use provided parameters).

type preset:

str | None

param filter_type:

Type of bandpass filter implementation:

‘efilt’: Complex frequency-shifted first-order lowpass (default). MATLAB AMT compatible. Creates resonant filters with asymmetric response.
‘butterworth’: True Butterworth bandpass with symmetric flat passband. Uses Second-Order Sections (SOS) for numerical stability. Conceptually more accurate but numerically different from AMT. Extension not present in original MATLAB AMT.

Default: 'efilt'.

type filter_type:

str

param learnable:

If True, make filter coefficients trainable parameters. If False, register them as buffers. Default: False.

type learnable:

bool

param dtype:

Data type for internal computations. Default: torch.float32.

type dtype:

dtype

fs

Sampling rate in Hz.

Type:: float

fc

Center frequencies of auditory channels, shape (n_channels,).

Type:: torch.Tensor

Q

Q-factor for modulation filters.

Type:: float

max_mfc

Maximum modulation frequency in Hz.

Type:: float

lp_cutoff

Cutoff frequency for main lowpass filter (mfc=0) in Hz.

Type:: float

att_factor

Attenuation factor for high modulation frequencies.

Type:: float

use_upper_limit

Whether dynamic upper limit is applied.

Type:: bool

preset

Selected preset configuration name.

Type:: str or None

filter_type

Bandpass filter implementation type (‘efilt’ or ‘butterworth’).

Type:: str

learnable

Whether filter coefficients are trainable.

Type:: bool

num_channels

Number of auditory frequency channels.

Type:: int

use_lp150_prefilter

Whether 150 Hz lowpass pre-filtering is enabled.

Type:: bool

b_lowpass

Numerator coefficients for main lowpass filter (mfc=0).

Type:: torch.Tensor

a_lowpass

Denominator coefficients for main lowpass filter (mfc=0).

Type:: torch.Tensor

b_lp150

Numerator coefficients for optional 150 Hz pre-filter.

Type:: torch.Tensor

a_lp150

Denominator coefficients for optional 150 Hz pre-filter.

Type:: torch.Tensor

mfc

Modulation center frequencies for each auditory channel. Length equals num_channels, each element is 1D tensor with variable length depending on the upper modulation frequency limit.

Type:: List[torch.Tensor]

filter_coeffs

Filter coefficients for all modulation filters, organized by channel.

Type:: nn.ModuleList

Shape

-----

- Input

\(B\) = batch size
\(F\) = frequency channels (length of fc)
\(T\) = time samples

Type:: \((B, F, T)\) where

- Output

\((B, M_f, T)\) and \(M_f\) is the number of modulation filters for frequency channel \(f\) (varies per channel when use_upper_limit=True).

Type:: List[torch.Tensor] of length \(F\), where each element has shape

Notes

Preset Differences:

Preset	lp_cutoff (Hz)	att_factor	use_upper_limit	150 Hz pre-filter
dau1997	2.5	1.0	False	No
jepsen2008	2.5	1/√2 (0.707)	True (0.25xfc)	Yes
paulick2024	2.5	1/√2 (0.707)	True (0.25xfc)	Yes

Filter Type Comparison:

efilt: MATLAB AMT compatible. Implements frequency-shifted lowpass, creating resonant bandpass filters. Not a true symmetric bandpass, but matches original implementations.
butterworth: Extension for improved numerical stability. True symmetric bandpass with flat passband. Uses SOS format to avoid coefficient overflow. This option is not available in original MATLAB AMT.

150 Hz Pre-filtering:

For jepsen2008 and paulick2024 presets, a 1st-order Butterworth lowpass at 150 Hz is applied before the modulation filterbank to remove very high modulation frequencies. This is motivated by physiological limitations and improves predictions for certain psychoacoustic tasks (Kohlrausch et al. 2000).

The 150 Hz pre-filter is separate from the main 2.5 Hz lowpass (mfc=0), which remains constant across all presets.

Computational Complexity:

Time complexity: \(O(B \cdot F \cdot M \cdot T)\) where \(M\) is the average number of modulation filters per channel
The number of modulation filters varies per frequency channel when use_upper_limit=True
For typical configurations: 8-15 modulation filters per channel

See also

IHCEnvelope: Inner hair cell envelope extraction (preprocessing)
AdaptLoop: Auditory nerve adaptation (typically follows modulation filterbank)
GammatoneFilterbank: Gammatone auditory filterbank (input source)
DRNLFilterbank: Dual resonance nonlinear filterbank (alternative input)
headphonefilter: Headphone/outer ear frequency response
middleearfilter: Middle ear transfer function

Examples

Basic usage with default parameters:

>>> import torch
>>> from torch_amt.common.modulation import ModulationFilterbank
>>>
>>> # Create auditory center frequencies (31 channels, ERB-spaced)
>>> fc = torch.linspace(100, 8000, 31)
>>>
>>> # Initialize modulation filterbank
>>> modfb = ModulationFilterbank(fs=16000, fc=fc)
>>>
>>> # Process filterbank output (2 batches, 31 channels, 1 sec)
>>> x = torch.randn(2, 31, 16000)
>>> y = modfb(x)
>>>
>>> print(f"Input: {x.shape}")
Input: torch.Size([2, 31, 16000])
>>> print(f"Output: List of {len(y)} tensors")
Output: List of 31 tensors
>>> print(f"First channel shape: {y[0].shape}")
First channel shape: torch.Size([2, 13, 16000])

Using preset configurations:

>>> # Dau et al. (1997): Original CASP model
>>> modfb_dau = ModulationFilterbank(fs=16000, fc=fc, preset='dau1997')
>>> print(f"Dau1997 - att_factor: {modfb_dau.att_factor:.3f}")
Dau1997 - att_factor: 1.000
>>>
>>> # Jepsen et al. (2008): Updated model with 150 Hz pre-filter
>>> modfb_jep = ModulationFilterbank(fs=16000, fc=fc, preset='jepsen2008')
>>> print(f"Jepsen2008 - att_factor: {modfb_jep.att_factor:.3f}")
Jepsen2008 - att_factor: 0.707
>>> print(f"150 Hz pre-filter: {modfb_jep.use_lp150_prefilter}")
150 Hz pre-filter: True

Comparing filter types:

>>> # Default: efilt (MATLAB compatible)
>>> modfb_efilt = ModulationFilterbank(fs=16000, fc=fc[:5], filter_type='efilt')
>>>
>>> # Alternative: Butterworth (improved stability)
>>> modfb_butter = ModulationFilterbank(fs=16000, fc=fc[:5], filter_type='butterworth')
>>>
>>> x_test = torch.randn(1, 5, 8000)
>>> y_efilt = modfb_efilt(x_test)
>>> y_butter = modfb_butter(x_test)
>>> # Outputs will differ numerically but capture similar modulation content

Learnable modulation filterbank for neural network training:

>>> modfb_learn = ModulationFilterbank(fs=16000, fc=fc[:10], learnable=True)
>>>
>>> # Check trainable parameters
>>> n_params = sum(p.numel() for p in modfb_learn.parameters())
>>> print(f"Trainable parameters: {n_params}")
Trainable parameters: ...
>>>
>>> # Use in training loop
>>> optimizer = torch.optim.Adam(modfb_learn.parameters(), lr=1e-3)
>>> # ... training code ...

References

__init__(fs, fc, Q=2.0, max_mfc=150.0, lp_cutoff=2.5, att_factor=1.0, use_upper_limit=False, preset=None, filter_type='efilt', learnable=False, dtype=torch.float32)[source]

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

Parameters:

fs (float)
fc (Tensor)
Q (float)
max_mfc (float)
lp_cutoff (float)
att_factor (float)
use_upper_limit (bool)
preset (str | None)
filter_type (str)
learnable (bool)
dtype (dtype)

forward(x)[source]

Apply modulation filterbank to input signal.

OPTIMIZED VERSION: Vectorized batch processing eliminates nested loops for significant performance improvement (~15-30x speedup).

Processes each auditory frequency channel independently through its corresponding modulation filterbank. Applies optional 150 Hz pre-filtering, then filters with lowpass (mfc=0) and bandpass modulation filters, and finally applies phase processing (real vs envelope extraction).

Parameters:

x (Tensor) –

Input signal from auditory filterbank (e.g., Gammatone, DRNL). Shape: \((B, F, T)\) or \((F, T)\) where:

\(B\) = batch size (optional)
\(F\) = frequency channels (must match length of self.fc)
\(T\) = time samples

Returns:

List of length \(F\) (one per frequency channel). Each element is a tensor with shape:

\((B, M_f, T)\) if input had batch dimension
\((M_f, T)\) if input was 2D

where \(M_f\) is the number of modulation filters for channel \(f\) (varies per channel when use_upper_limit=True).

Return type:

List[Tensor]

Notes

Processing steps per channel:

Optional 150 Hz lowpass pre-filtering (if use_lp150_prefilter=True)
Lowpass filter (mfc=0, cutoff 2.5 Hz)
Bandpass filters (mfc > 0) with 6 dB emphasis
Phase processing:
- mfc ≤ 10 Hz: Real part (preserves phase)
- mfc > 10 Hz: Envelope (|·|) with attenuation factor

The output list structure (vs single tensor) allows variable number of modulation filters per channel, which occurs when use_upper_limit=True.

Optimization Strategy:

All batch samples processed in parallel (no batch loop)
Filters applied to all batches simultaneously using vectorized operations
JIT-compiled IIR filtering for time-sequential processing

extra_repr()[source]

Extra representation string for module printing.

Returns:: String containing key module parameters.
Return type:: str

T_destination = ~T_destination

add_module(name, module)

Add a child module to the current module.

The module can be accessed as an attribute using the given name.

Args:

name (str): name of the child module. The child module can be: accessed from this module using the given name

module (Module): child module to be added to the module.

Parameters:

name (str)
module (Module | None)

Return type:

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

Typical use includes initializing the parameters of a model (see also torch.nn.init).

Args:: fn (Module -> None): function to be applied to each submodule
Returns:: Module: self

Example:

>>> @torch.no_grad()
>>> def init_weights(m):
>>>     print(m)
>>>     if type(m) is nn.Linear:
>>>         m.weight.fill_(1.0)
>>>         print(m.weight)
>>> net = nn.Sequential(nn.Linear(2, 2), nn.Linear(2, 2))
>>> net.apply(init_weights)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)

Parameters:: fn (Callable[[Module], None])
Return type:: Self

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

buffers(recurse=True)

Return an iterator over module buffers.

Args:

recurse (bool): if True, then yields buffers of this module: and all submodules. Otherwise, yields only buffers that are direct members of this module.

Yields:

torch.Tensor: module buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for buf in model.buffers():
>>>     print(type(buf), buf.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Tensor]

call_super_init: bool = False

children()

Return an iterator over immediate children modules.

Return type:: Iterator[Module]

Yields:: Module: a child module

compile(*args, **kwargs)

Compile this Module’s forward using torch.compile().

This Module’s __call__ method is compiled and all arguments are passed as-is to torch.compile().

See torch.compile() for details on the arguments for this function.

Return type:: None

cpu()

Move all model parameters and buffers to the CPU.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

cuda(device=None)

Move all model parameters and buffers to the GPU.

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

Note

This method modifies the module in-place.

Args:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

double()

Casts all floating point parameters and buffers to double datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

dump_patches: bool = False

eval()

Set the module in evaluation mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e. whether 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.

Return type:: Self

Returns:: Module: self

float()

Casts all floating point parameters and buffers to float datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the buffer: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.Tensor: The buffer referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not a buffer

Parameters:: target (str)
Return type:: Tensor

get_extra_state()

Return 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 picklable to ensure working serialization of the state_dict. We only provide backwards compatibility guarantees for serializing Tensors; other objects may break backwards compatibility if their serialized pickled form changes.

Return type:: Any

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

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the Parameter: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.nn.Parameter: The Parameter referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not an nn.Parameter

Parameters:: target (str)
Return type:: Parameter

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(16, 33, kernel_size=(3, 3), stride=(2, 2))
        )
        (linear): Linear(in_features=100, out_features=200, bias=True)
    )
)

(The diagram shows an nn.Module A. A which 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.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

Returns:

torch.nn.Module: The submodule referenced by target

Raises:

AttributeError: If at any point along the path resulting from: the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:: target (str)
Return type:: Module

half()

Casts all floating point parameters and buffers to half datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

ipu(device=None)

Move all model parameters and buffers to the IPU.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

load_state_dict(state_dict, strict=True, assign=False)

Copy parameters and buffers from state_dict into this module and its descendants.

If strict is True, then the keys of state_dict must exactly match the keys returned by this module’s state_dict() function.

Warning

If assign is True the optimizer must be created after the call to load_state_dict unless get_swap_module_params_on_conversion() is True.

Args:

state_dict (dict): a dict containing parameters and: persistent buffers.
strict (bool, optional): whether to strictly enforce that the keys: in state_dict match the keys returned by this module’s state_dict() function. Default: True
assign (bool, optional): When set to False, the properties of the tensors: in the current module are preserved whereas setting it to True preserves properties of the Tensors in the state dict. The only exception is the requires_grad field of Parameter for which the value from the module is preserved. Default: False

Returns:

NamedTuple with missing_keys and unexpected_keys fields:

missing_keys is a list of str containing any keys that are expected
by this module but missing from the provided state_dict.
unexpected_keys is a list of str containing the keys that are not
expected by this module but present in the provided state_dict.

Note:

If a parameter or buffer is registered as None and its corresponding key exists in state_dict, load_state_dict() will raise a RuntimeError.

Parameters:

state_dict (Mapping[str, Any])
strict (bool)
assign (bool)

modules()

Return an iterator over all modules in the network.

Return type:: Iterator[Module]

Yields:: Module: a module in the network
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.modules()):
...     print(idx, '->', m)

0 -> Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)
1 -> Linear(in_features=2, out_features=2, bias=True)

mtia(device=None)

Move all model parameters and buffers to the MTIA.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

named_buffers(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

Args:

prefix (str): prefix to prepend to all buffer names. recurse (bool, optional): if True, then yields buffers of this module

and all submodules. Otherwise, yields only buffers that are direct members of this module. Defaults to True.

remove_duplicate (bool, optional): whether to remove the duplicated buffers in the result. Defaults to True.

Yields:

(str, torch.Tensor): Tuple containing the name and buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, buf in self.named_buffers():
>>>     if name in ['running_var']:
>>>         print(buf.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

Return type:: Iterator[tuple[str, Module]]

Yields:: (str, Module): Tuple containing a name and child module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, module in model.named_children():
>>>     if name in ['conv4', 'conv5']:
>>>         print(module)

named_modules(memo=None, prefix='', remove_duplicate=True)

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

Args:: memo: a memo to store the set of modules already added to the result prefix: a prefix that will be added to the name of the module remove_duplicate: whether to remove the duplicated module instances in the result

or not
Yields:: (str, Module): Tuple of name and module
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.named_modules()):
...     print(idx, '->', m)

0 -> ('', Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
))
1 -> ('0', Linear(in_features=2, out_features=2, bias=True))

Parameters:

memo (set[Module] | None)
prefix (str)
remove_duplicate (bool)

named_parameters(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

Args:

prefix (str): prefix to prepend to all parameter names. recurse (bool): if True, then yields parameters of this module

and all submodules. Otherwise, yields only parameters that are direct members of this module.

remove_duplicate (bool, optional): whether to remove the duplicated: parameters in the result. Defaults to True.

Yields:

(str, Parameter): Tuple containing the name and parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, param in self.named_parameters():
>>>     if name in ['bias']:
>>>         print(param.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

parameters(recurse=True)

Return an iterator over module parameters.

This is typically passed to an optimizer.

Args:

recurse (bool): if True, then yields parameters of this module: and all submodules. Otherwise, yields only parameters that are direct members of this module.

Yields:

Parameter: module parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for param in model.parameters():
>>>     print(type(param), param.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Parameter]

register_backward_hook(hook)

Register 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:

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

Parameters:: hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
Return type:: RemovableHandle

register_buffer(name, tensor, persistent=True)

Add a buffer to the module.

This is typically used to register a buffer that should not be considered a model parameter. For example, BatchNorm’s running_mean is not a parameter, but is part of the module’s state. Buffers, by default, are persistent and will be saved alongside parameters. This behavior can be changed by setting persistent to False. The only difference between a persistent buffer and a non-persistent buffer is that the latter will not be a part of this module’s state_dict.

Buffers can be accessed as attributes using given names.

Args:

name (str): name of the buffer. The buffer can be accessed: from this module using the given name
tensor (Tensor or None): buffer to be registered. If None, then operations: that run on buffers, such as cuda, are ignored. If None, the buffer is not included in the module’s state_dict.
persistent (bool): whether the buffer is part of this module’s: state_dict.

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> self.register_buffer('running_mean', torch.zeros(num_features))

Parameters:

name (str)
tensor (Tensor | None)
persistent (bool)

Return type:

register_forward_hook(hook, *, prepend=False, with_kwargs=False, always_call=False)

Register a forward hook on the module.

The hook will be called every time after forward() has computed an output.

If with_kwargs is False or not specified, 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. The hook should have the following signature:

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

If with_kwargs is True, the forward hook will be passed the kwargs given to the forward function and be expected to return the output possibly modified. The hook should have the following signature:

hook(module, args, kwargs, output) -> None or modified output

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If True, the provided hook will be fired

before all existing forward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward hooks on this torch.nn.Module. Note that global forward hooks registered with register_module_forward_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If True, the hook will be passed the: kwargs given to the forward function. Default: False
always_call (bool): If True the hook will be run regardless of: whether an exception is raised while calling the Module. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...], Any], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any], Any], Any | None])
prepend (bool)
with_kwargs (bool)
always_call (bool)

Return type:

RemovableHandle

register_forward_pre_hook(hook, *, prepend=False, with_kwargs=False)

Register a forward pre-hook on the module.

The hook will be called every time before forward() is invoked.

If with_kwargs is false or not specified, 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). The hook should have the following signature:

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

If with_kwargs is true, the forward pre-hook will be passed the kwargs given to the forward function. And if the hook modifies the input, both the args and kwargs should be returned. The hook should have the following signature:

hook(module, args, kwargs) -> None or a tuple of modified input and kwargs

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing forward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward_pre hooks on this torch.nn.Module. Note that global forward_pre hooks registered with register_module_forward_pre_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If true, the hook will be passed the kwargs: given to the forward function. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...]], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any]], tuple[Any, dict[str, Any]] | None])
prepend (bool)
with_kwargs (bool)

Return type:

RemovableHandle

register_full_backward_hook(hook, prepend=False)

Register a backward hook on the module.

The hook will be called every time the gradients with respect to a module are computed, and its firing rules are as follows:

Ordinarily, the hook fires when the gradients are computed with respect to the module inputs.

If none of the module inputs require gradients, the hook will fire when the gradients are computed with respect to module outputs.

If none of the module outputs require gradients, then the hooks will not fire.

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.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward hooks on this torch.nn.Module. Note that global backward hooks registered with register_module_full_backward_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_full_backward_pre_hook(hook, prepend=False)

Register a backward pre-hook on the module.

The hook will be called every time the gradients for the module are computed. The hook should have the following signature:

hook(module, grad_output) -> tuple[Tensor, ...], Tensor or None

The grad_output is a tuple. The hook should not modify its arguments, but it can optionally return a new gradient with respect to the output that will be used in place of grad_output in subsequent computations. Entries in 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 inplace is not allowed when using backward hooks and will raise an error.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward_pre hooks on this torch.nn.Module. Note that global backward_pre hooks registered with register_module_full_backward_pre_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module’s load_state_dict() is called.

It should have the following signature::: hook(module, incompatible_keys) -> None

The module argument is the current module that this hook is registered on, and the incompatible_keys argument is a NamedTuple consisting of attributes missing_keys and unexpected_keys. missing_keys is a list of str containing the missing keys and unexpected_keys is a list of str containing the unexpected keys.

The given incompatible_keys can be modified inplace if needed.

Note that the checks performed when calling load_state_dict() with strict=True are affected by modifications the hook makes to missing_keys or unexpected_keys, as expected. Additions to either set of keys will result in an error being thrown when strict=True, and clearing out both missing and unexpected keys will avoid an error.

Returns:

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

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module’s load_state_dict() is called.

It should have the following signature::

hook(module, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) -> None # noqa: B950

Arguments:

hook (Callable): Callable hook that will be invoked before: loading the state dict.

register_module(name, module)

Alias for add_module().

Parameters:

name (str)
module (Module | None)

Return type:

register_parameter(name, param)

Add a parameter to the module.

The parameter can be accessed as an attribute using given name.

Args:

name (str): name of the parameter. The parameter can be accessed: from this module using the given name
param (Parameter or None): parameter to be added to the module. If: None, then operations that run on parameters, such as cuda, are ignored. If None, the parameter is not included in the module’s state_dict.

Parameters:

name (str)
param (Parameter | None)

Return type:

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

It should have the following signature::: hook(module, state_dict, prefix, local_metadata) -> None

The registered hooks can modify the state_dict inplace.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

It should have the following signature::: hook(module, prefix, keep_vars) -> None

The registered hooks can be used to perform pre-processing before the state_dict call is made.

requires_grad_(requires_grad=True)

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.

Args:

requires_grad (bool): whether autograd should record operations on: parameters in this module. Default: True.

Returns:

Module: self

Parameters:: requires_grad (bool)
Return type:: Self

set_extra_state(state)

Set extra state contained in the loaded state_dict.

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.

Args:: state (dict): Extra state from the state_dict

Parameters:: state (Any)
Return type:: None

set_submodule(target, module, strict=False)

Set the submodule given by target if it exists, otherwise throw an error.

Note

If strict is set to False (default), the method will replace an existing submodule or create a new submodule if the parent module exists. If strict is set to True, the method will only attempt to replace an existing submodule and throw an error if the submodule does not exist.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(3, 3, 3)
        )
        (linear): Linear(3, 3)
    )
)

(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 override the Conv2d with a new submodule Linear, you could call set_submodule("net_b.net_c.conv", nn.Linear(1, 1)) where strict could be True or False

To add a new submodule Conv2d to the existing net_b module, you would call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1)).

In the above if you set strict=True and call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1), strict=True), an AttributeError will be raised because net_b does not have a submodule named conv.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

module: The module to set the submodule to. strict: If False, the method will replace an existing submodule

or create a new submodule if the parent module exists. If True, the method will only attempt to replace an existing submodule and throw an error if the submodule doesn’t already exist.

Raises:

ValueError: If the target string is empty or if module is not an instance of nn.Module. AttributeError: If at any point along the path resulting from

the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:

target (str)
module (Module)
strict (bool)

Return type:

share_memory()

Return type:: Self

state_dict(*args, destination=None, prefix='', keep_vars=False)

Overloads:

self, destination (T_destination), prefix (str), keep_vars (bool) → T_destination
self, prefix (str), keep_vars (bool) → dict[str, Any]

Return a dictionary containing references to the whole state of the module.

Both parameters and persistent buffers (e.g. running averages) are included. Keys are corresponding parameter and buffer names. Parameters and buffers set to None are not included.

Note

The returned object is a shallow copy. It contains references to the module’s parameters and buffers.

Warning

Currently state_dict() also accepts positional arguments for destination, prefix and keep_vars in order. However, this is being deprecated and keyword arguments will be enforced in future releases.

Warning

Please avoid the use of argument destination as it is not designed for end-users.

Args:

destination (dict, optional): If provided, the state of module will: be updated into the dict and the same object is returned. Otherwise, an OrderedDict will be created and returned. Default: None.
prefix (str, optional): a prefix added to parameter and buffer: names to compose the keys in state_dict. Default: ''.
keep_vars (bool, optional): by default the Tensor s: returned in the state dict are detached from autograd. If it’s set to True, detaching will not be performed. Default: False.

Returns:

dict:: a dictionary containing a whole state of the module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> module.state_dict().keys()
['bias', 'weight']

to(*args, **kwargs)

Overloads:

self, device (DeviceLikeType | None), dtype (dtype | None), non_blocking (bool) → Self
self, dtype (dtype), non_blocking (bool) → Self
self, tensor (Tensor), non_blocking (bool) → Self

Move and/or cast 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.

Args:

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:

Module: self

Examples:

>>> # xdoctest: +IGNORE_WANT("non-deterministic")
>>> 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)
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA1)
>>> 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, recurse=True)

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

Args:

device (torch.device): The desired device of the parameters: and buffers in this module.
recurse (bool): Whether parameters and buffers of submodules should: be recursively moved to the specified device.

Returns:

Module: self

Parameters:

device (str | device | int | None)
recurse (bool)

Return type:

Self

train(mode=True)

Set the module in training mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e., whether they are affected, e.g. Dropout, BatchNorm, etc.

Args:

mode (bool): whether to set training mode (True) or evaluation: mode (False). Default: True.

Returns:

Module: self

Parameters:: mode (bool)
Return type:: Self

type(dst_type)

Casts all parameters and buffers to dst_type.

Note

This method modifies the module in-place.

Args:: dst_type (type or string): the desired type
Returns:: Module: self

Parameters:: dst_type (dtype | str)
Return type:: Self

xpu(device=None)

Move 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.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

zero_grad(set_to_none=True)

Reset gradients of all model parameters.

See similar function under torch.optim.Optimizer for more context.

Args:

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

Parameters:: set_to_none (bool)
Return type:: None

training: bool

FastModulationFilterbank

class torch_amt.FastModulationFilterbank(*args, tolerance=2, **kwargs)[source]

Bases: ModulationFilterbank

Optimized modulation filterbank using mega-batch processing.

Drop-in replacement for ModulationFilterbank that achieves 3-5x speedup through vectorized mega-batch processing while maintaining numerical accuracy (difference < 1e-12). Inherits all functionality from parent class but overrides the forward pass to use intelligent batching of filter operations.

Key Optimizations:

Mega-batch flattening: Combines all channel-filter pairs into single batch
Filter grouping: Groups identical coefficients for parallel processing
Vectorized operations: Eliminates Python loops in favor of tensor ops
Intelligent reconstruction: Uses pre-built indexing for ragged outputs

Performance:

Speedup: 3-5x faster than standard ModulationFilterbank
Accuracy: Numerically identical (max diff < 1e-12)
Memory: Same as parent class (no overhead)

Parameters:

tolerance (int) – Number of decimal places for coefficient rounding during filter grouping. Lower values create more groups (slower but more accurate). Higher values create fewer groups (faster but less accurate). Default: 2 (optimal tradeoff: 34x speedup within grouping, max_diff=1.69e-3).
*args – All other parameters inherited from ModulationFilterbank. See parent class documentation for complete parameter list.
**kwargs – All other parameters inherited from ModulationFilterbank. See parent class documentation for complete parameter list.

tolerance

Coefficient rounding tolerance for filter grouping.

Type:: int

megabatch_map

Pre-computed mapping structure for flattening/reconstruction. Each entry contains channel index, filter index, and modulation frequency.

Type:: list

total_filters

Total number of channel-filter combinations across all channels.

Type:: int

Inherits all attributes from :class:`ModulationFilterbank`.

Shape

-----

- Input

\((B, F, T)\) where \(B\) = batch size, \(F\) = frequency channels, \(T\) = time samples

Type:: Same as ModulationFilterbank

- Output

List[torch.Tensor] of length \(F\), where each element has shape \((B, M_f, T)\) and \(M_f\) varies per channel

Type:: Same as ModulationFilterbank

Notes

Filter Grouping Strategy:

The tolerance parameter controls filter coefficient grouping:

tolerance=1: More groups, slower, high accuracy (max_diff ~ 1e-4)
tolerance=2: Balanced (default), 34x speedup, max_diff ~ 1.7e-3
tolerance=3: Fewer groups, fastest, lower accuracy (max_diff ~ 0.01)

Grouping is performed each forward pass when learnable=True to adapt to parameter changes. For fixed filters (learnable=False), grouping could be cached but current implementation recomputes for simplicity.

Gradient Flow for Learnable Filters:

When learnable=True, filter coefficients can be trained via gradient descent. The mega-batch processing preserves full gradient flow, but filter grouping means only one filter per group directly receives gradients during forward pass.

To distribute gradients to all filters in each group, call distribute_gradients() after backward() and before optimizer.step():

output = fast_mod(x)
loss = criterion(output, target)
loss.backward()
fast_mod.distribute_gradients()  # ← Important for learnable filters!
optimizer.step()

Computational Complexity:

Time: \(O(B \cdot F \cdot M \cdot T / G)\) where \(G\) is number of filter groups (typically 5-10x fewer than total filters)
Space: \(O(B \cdot F \cdot M \cdot T)\) same as parent

See also

ModulationFilterbank: Standard implementation (reference)
FastKing2019ModulationFilterbank: FFT-based fast version for King2019

Examples

Drop-in replacement for standard filterbank:

>>> import torch
>>> from torch_amt.common.modulation import FastModulationFilterbank
>>>
>>> # Create auditory center frequencies
>>> fc = torch.linspace(100, 8000, 31)
>>>
>>> # Initialize fast modulation filterbank
>>> modfb = FastModulationFilterbank(fs=16000, fc=fc, preset='jepsen2008')
>>>
>>> # Same API as ModulationFilterbank
>>> x = torch.randn(2, 31, 16000)
>>> y = modfb(x)  # 3-5x faster!
>>>
>>> print(f\"Output: List of {len(y)} tensors\")
Output: List of 31 tensors
>>> print(f\"First channel shape: {y[0].shape}\")
First channel shape: torch.Size([2, 13, 16000])

Training with learnable filters:

>>> # Create learnable filterbank
>>> modfb_learn = FastModulationFilterbank(
...     fs=16000, fc=fc[:10], preset='dau1997', learnable=True
... )
>>>
>>> # Training loop
>>> optimizer = torch.optim.Adam(modfb_learn.parameters(), lr=1e-3)
>>>
>>> for epoch in range(10):
...     output = modfb_learn(x[:, :10, :])
...     loss = criterion(output, target)
...     loss.backward()
...     modfb_learn.distribute_gradients()  # ← Essential!
...     optimizer.step()
...     optimizer.zero_grad()

Adjusting tolerance for speed/accuracy tradeoff:

>>> # High accuracy (more groups, slower)
>>> modfb_precise = FastModulationFilterbank(fs=16000, fc=fc, tolerance=1)
>>>
>>> # Balanced (default)
>>> modfb_balanced = Fast ModulationFilterbank(fs=16000, fc=fc, tolerance=2)
>>>
>>> # Maximum speed (fewer groups, minor accuracy loss)
>>> modfb_fast = FastModulationFilterbank(fs=16000, fc=fc, tolerance=3)

References

__init__(*args, tolerance=2, **kwargs)[source]

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

Parameters:: tolerance (int)

forward(x)[source]

Apply modulation filterbank using mega-batch processing.

This is 3-5x faster than the standard implementation while producing numerically identical results.

Parameters:: x (Tensor) – Input signal, shape (B, C, T) or (C, T)
Returns:: List of tensors, one per channel, same as ModulationFilterbank
Return type:: list

distribute_gradients()[source]

Distribute gradients from representative filters to all group members.

Call this method after backward() to ensure all filter coefficients receive gradient updates, even if they weren’t directly used in the forward pass due to filter grouping.

This implements weight sharing where filters in the same group receive the same gradient update.

Example

>>> fast_mod = FastModulationFilterbank(fs=44100, fc=fc, learnable=True)
>>> output = fast_mod(x)
>>> loss = compute_loss(output)
>>> loss.backward()
>>> fast_mod.distribute_gradients()  # ← Call this!
>>> optimizer.step()

T_destination = ~T_destination

add_module(name, module)

Add a child module to the current module.

The module can be accessed as an attribute using the given name.

Args:

name (str): name of the child module. The child module can be: accessed from this module using the given name

module (Module): child module to be added to the module.

Parameters:

name (str)
module (Module | None)

Return type:

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

Typical use includes initializing the parameters of a model (see also torch.nn.init).

Args:: fn (Module -> None): function to be applied to each submodule
Returns:: Module: self

Example:

>>> @torch.no_grad()
>>> def init_weights(m):
>>>     print(m)
>>>     if type(m) is nn.Linear:
>>>         m.weight.fill_(1.0)
>>>         print(m.weight)
>>> net = nn.Sequential(nn.Linear(2, 2), nn.Linear(2, 2))
>>> net.apply(init_weights)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)

Parameters:: fn (Callable[[Module], None])
Return type:: Self

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

buffers(recurse=True)

Return an iterator over module buffers.

Args:

recurse (bool): if True, then yields buffers of this module: and all submodules. Otherwise, yields only buffers that are direct members of this module.

Yields:

torch.Tensor: module buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for buf in model.buffers():
>>>     print(type(buf), buf.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Tensor]

call_super_init: bool = False

children()

Return an iterator over immediate children modules.

Return type:: Iterator[Module]

Yields:: Module: a child module

compile(*args, **kwargs)

Compile this Module’s forward using torch.compile().

This Module’s __call__ method is compiled and all arguments are passed as-is to torch.compile().

See torch.compile() for details on the arguments for this function.

Return type:: None

cpu()

Move all model parameters and buffers to the CPU.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

cuda(device=None)

Move all model parameters and buffers to the GPU.

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

Note

This method modifies the module in-place.

Args:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

double()

Casts all floating point parameters and buffers to double datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

dump_patches: bool = False

eval()

Set the module in evaluation mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e. whether 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.

Return type:: Self

Returns:: Module: self

extra_repr()

Extra representation string for module printing.

Returns:: String containing key module parameters.
Return type:: str

float()

Casts all floating point parameters and buffers to float datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the buffer: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.Tensor: The buffer referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not a buffer

Parameters:: target (str)
Return type:: Tensor

get_extra_state()

Return 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 picklable to ensure working serialization of the state_dict. We only provide backwards compatibility guarantees for serializing Tensors; other objects may break backwards compatibility if their serialized pickled form changes.

Return type:: Any

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

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the Parameter: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.nn.Parameter: The Parameter referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not an nn.Parameter

Parameters:: target (str)
Return type:: Parameter

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(16, 33, kernel_size=(3, 3), stride=(2, 2))
        )
        (linear): Linear(in_features=100, out_features=200, bias=True)
    )
)

(The diagram shows an nn.Module A. A which 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.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

Returns:

torch.nn.Module: The submodule referenced by target

Raises:

AttributeError: If at any point along the path resulting from: the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:: target (str)
Return type:: Module

half()

Casts all floating point parameters and buffers to half datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

ipu(device=None)

Move all model parameters and buffers to the IPU.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

load_state_dict(state_dict, strict=True, assign=False)

Copy parameters and buffers from state_dict into this module and its descendants.

If strict is True, then the keys of state_dict must exactly match the keys returned by this module’s state_dict() function.

Warning

If assign is True the optimizer must be created after the call to load_state_dict unless get_swap_module_params_on_conversion() is True.

Args:

state_dict (dict): a dict containing parameters and: persistent buffers.
strict (bool, optional): whether to strictly enforce that the keys: in state_dict match the keys returned by this module’s state_dict() function. Default: True
assign (bool, optional): When set to False, the properties of the tensors: in the current module are preserved whereas setting it to True preserves properties of the Tensors in the state dict. The only exception is the requires_grad field of Parameter for which the value from the module is preserved. Default: False

Returns:

NamedTuple with missing_keys and unexpected_keys fields:

missing_keys is a list of str containing any keys that are expected
by this module but missing from the provided state_dict.
unexpected_keys is a list of str containing the keys that are not
expected by this module but present in the provided state_dict.

Note:

If a parameter or buffer is registered as None and its corresponding key exists in state_dict, load_state_dict() will raise a RuntimeError.

Parameters:

state_dict (Mapping[str, Any])
strict (bool)
assign (bool)

modules()

Return an iterator over all modules in the network.

Return type:: Iterator[Module]

Yields:: Module: a module in the network
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.modules()):
...     print(idx, '->', m)

0 -> Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)
1 -> Linear(in_features=2, out_features=2, bias=True)

mtia(device=None)

Move all model parameters and buffers to the MTIA.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

named_buffers(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

Args:

prefix (str): prefix to prepend to all buffer names. recurse (bool, optional): if True, then yields buffers of this module

and all submodules. Otherwise, yields only buffers that are direct members of this module. Defaults to True.

remove_duplicate (bool, optional): whether to remove the duplicated buffers in the result. Defaults to True.

Yields:

(str, torch.Tensor): Tuple containing the name and buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, buf in self.named_buffers():
>>>     if name in ['running_var']:
>>>         print(buf.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

Return type:: Iterator[tuple[str, Module]]

Yields:: (str, Module): Tuple containing a name and child module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, module in model.named_children():
>>>     if name in ['conv4', 'conv5']:
>>>         print(module)

named_modules(memo=None, prefix='', remove_duplicate=True)

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

Args:: memo: a memo to store the set of modules already added to the result prefix: a prefix that will be added to the name of the module remove_duplicate: whether to remove the duplicated module instances in the result

or not
Yields:: (str, Module): Tuple of name and module
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.named_modules()):
...     print(idx, '->', m)

0 -> ('', Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
))
1 -> ('0', Linear(in_features=2, out_features=2, bias=True))

Parameters:

memo (set[Module] | None)
prefix (str)
remove_duplicate (bool)

named_parameters(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

Args:

prefix (str): prefix to prepend to all parameter names. recurse (bool): if True, then yields parameters of this module

and all submodules. Otherwise, yields only parameters that are direct members of this module.

remove_duplicate (bool, optional): whether to remove the duplicated: parameters in the result. Defaults to True.

Yields:

(str, Parameter): Tuple containing the name and parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, param in self.named_parameters():
>>>     if name in ['bias']:
>>>         print(param.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

parameters(recurse=True)

Return an iterator over module parameters.

This is typically passed to an optimizer.

Args:

recurse (bool): if True, then yields parameters of this module: and all submodules. Otherwise, yields only parameters that are direct members of this module.

Yields:

Parameter: module parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for param in model.parameters():
>>>     print(type(param), param.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Parameter]

register_backward_hook(hook)

Register 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:

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

Parameters:: hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
Return type:: RemovableHandle

register_buffer(name, tensor, persistent=True)

Add a buffer to the module.

This is typically used to register a buffer that should not be considered a model parameter. For example, BatchNorm’s running_mean is not a parameter, but is part of the module’s state. Buffers, by default, are persistent and will be saved alongside parameters. This behavior can be changed by setting persistent to False. The only difference between a persistent buffer and a non-persistent buffer is that the latter will not be a part of this module’s state_dict.

Buffers can be accessed as attributes using given names.

Args:

name (str): name of the buffer. The buffer can be accessed: from this module using the given name
tensor (Tensor or None): buffer to be registered. If None, then operations: that run on buffers, such as cuda, are ignored. If None, the buffer is not included in the module’s state_dict.
persistent (bool): whether the buffer is part of this module’s: state_dict.

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> self.register_buffer('running_mean', torch.zeros(num_features))

Parameters:

name (str)
tensor (Tensor | None)
persistent (bool)

Return type:

register_forward_hook(hook, *, prepend=False, with_kwargs=False, always_call=False)

Register a forward hook on the module.

The hook will be called every time after forward() has computed an output.

If with_kwargs is False or not specified, 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. The hook should have the following signature:

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

If with_kwargs is True, the forward hook will be passed the kwargs given to the forward function and be expected to return the output possibly modified. The hook should have the following signature:

hook(module, args, kwargs, output) -> None or modified output

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If True, the provided hook will be fired

before all existing forward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward hooks on this torch.nn.Module. Note that global forward hooks registered with register_module_forward_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If True, the hook will be passed the: kwargs given to the forward function. Default: False
always_call (bool): If True the hook will be run regardless of: whether an exception is raised while calling the Module. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...], Any], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any], Any], Any | None])
prepend (bool)
with_kwargs (bool)
always_call (bool)

Return type:

RemovableHandle

register_forward_pre_hook(hook, *, prepend=False, with_kwargs=False)

Register a forward pre-hook on the module.

The hook will be called every time before forward() is invoked.

If with_kwargs is false or not specified, 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). The hook should have the following signature:

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

If with_kwargs is true, the forward pre-hook will be passed the kwargs given to the forward function. And if the hook modifies the input, both the args and kwargs should be returned. The hook should have the following signature:

hook(module, args, kwargs) -> None or a tuple of modified input and kwargs

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing forward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward_pre hooks on this torch.nn.Module. Note that global forward_pre hooks registered with register_module_forward_pre_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If true, the hook will be passed the kwargs: given to the forward function. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...]], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any]], tuple[Any, dict[str, Any]] | None])
prepend (bool)
with_kwargs (bool)

Return type:

RemovableHandle

register_full_backward_hook(hook, prepend=False)

Register a backward hook on the module.

The hook will be called every time the gradients with respect to a module are computed, and its firing rules are as follows:

Ordinarily, the hook fires when the gradients are computed with respect to the module inputs.

If none of the module inputs require gradients, the hook will fire when the gradients are computed with respect to module outputs.

If none of the module outputs require gradients, then the hooks will not fire.

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.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward hooks on this torch.nn.Module. Note that global backward hooks registered with register_module_full_backward_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_full_backward_pre_hook(hook, prepend=False)

Register a backward pre-hook on the module.

The hook will be called every time the gradients for the module are computed. The hook should have the following signature:

hook(module, grad_output) -> tuple[Tensor, ...], Tensor or None

The grad_output is a tuple. The hook should not modify its arguments, but it can optionally return a new gradient with respect to the output that will be used in place of grad_output in subsequent computations. Entries in 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 inplace is not allowed when using backward hooks and will raise an error.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward_pre hooks on this torch.nn.Module. Note that global backward_pre hooks registered with register_module_full_backward_pre_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module’s load_state_dict() is called.

It should have the following signature::: hook(module, incompatible_keys) -> None

The module argument is the current module that this hook is registered on, and the incompatible_keys argument is a NamedTuple consisting of attributes missing_keys and unexpected_keys. missing_keys is a list of str containing the missing keys and unexpected_keys is a list of str containing the unexpected keys.

The given incompatible_keys can be modified inplace if needed.

Note that the checks performed when calling load_state_dict() with strict=True are affected by modifications the hook makes to missing_keys or unexpected_keys, as expected. Additions to either set of keys will result in an error being thrown when strict=True, and clearing out both missing and unexpected keys will avoid an error.

Returns:

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

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module’s load_state_dict() is called.

It should have the following signature::

hook(module, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) -> None # noqa: B950

Arguments:

hook (Callable): Callable hook that will be invoked before: loading the state dict.

register_module(name, module)

Alias for add_module().

Parameters:

name (str)
module (Module | None)

Return type:

register_parameter(name, param)

Add a parameter to the module.

The parameter can be accessed as an attribute using given name.

Args:

name (str): name of the parameter. The parameter can be accessed: from this module using the given name
param (Parameter or None): parameter to be added to the module. If: None, then operations that run on parameters, such as cuda, are ignored. If None, the parameter is not included in the module’s state_dict.

Parameters:

name (str)
param (Parameter | None)

Return type:

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

It should have the following signature::: hook(module, state_dict, prefix, local_metadata) -> None

The registered hooks can modify the state_dict inplace.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

It should have the following signature::: hook(module, prefix, keep_vars) -> None

The registered hooks can be used to perform pre-processing before the state_dict call is made.

requires_grad_(requires_grad=True)

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.

Args:

requires_grad (bool): whether autograd should record operations on: parameters in this module. Default: True.

Returns:

Module: self

Parameters:: requires_grad (bool)
Return type:: Self

set_extra_state(state)

Set extra state contained in the loaded state_dict.

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.

Args:: state (dict): Extra state from the state_dict

Parameters:: state (Any)
Return type:: None

set_submodule(target, module, strict=False)

Set the submodule given by target if it exists, otherwise throw an error.

Note

If strict is set to False (default), the method will replace an existing submodule or create a new submodule if the parent module exists. If strict is set to True, the method will only attempt to replace an existing submodule and throw an error if the submodule does not exist.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(3, 3, 3)
        )
        (linear): Linear(3, 3)
    )
)

(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 override the Conv2d with a new submodule Linear, you could call set_submodule("net_b.net_c.conv", nn.Linear(1, 1)) where strict could be True or False

To add a new submodule Conv2d to the existing net_b module, you would call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1)).

In the above if you set strict=True and call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1), strict=True), an AttributeError will be raised because net_b does not have a submodule named conv.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

module: The module to set the submodule to. strict: If False, the method will replace an existing submodule

or create a new submodule if the parent module exists. If True, the method will only attempt to replace an existing submodule and throw an error if the submodule doesn’t already exist.

Raises:

ValueError: If the target string is empty or if module is not an instance of nn.Module. AttributeError: If at any point along the path resulting from

the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:

target (str)
module (Module)
strict (bool)

Return type:

share_memory()

Return type:: Self

state_dict(*args, destination=None, prefix='', keep_vars=False)

Overloads:

self, destination (T_destination), prefix (str), keep_vars (bool) → T_destination
self, prefix (str), keep_vars (bool) → dict[str, Any]

Return a dictionary containing references to the whole state of the module.

Both parameters and persistent buffers (e.g. running averages) are included. Keys are corresponding parameter and buffer names. Parameters and buffers set to None are not included.

Note

The returned object is a shallow copy. It contains references to the module’s parameters and buffers.

Warning

Currently state_dict() also accepts positional arguments for destination, prefix and keep_vars in order. However, this is being deprecated and keyword arguments will be enforced in future releases.

Warning

Please avoid the use of argument destination as it is not designed for end-users.

Args:

destination (dict, optional): If provided, the state of module will: be updated into the dict and the same object is returned. Otherwise, an OrderedDict will be created and returned. Default: None.
prefix (str, optional): a prefix added to parameter and buffer: names to compose the keys in state_dict. Default: ''.
keep_vars (bool, optional): by default the Tensor s: returned in the state dict are detached from autograd. If it’s set to True, detaching will not be performed. Default: False.

Returns:

dict:: a dictionary containing a whole state of the module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> module.state_dict().keys()
['bias', 'weight']

to(*args, **kwargs)

Overloads:

self, device (DeviceLikeType | None), dtype (dtype | None), non_blocking (bool) → Self
self, dtype (dtype), non_blocking (bool) → Self
self, tensor (Tensor), non_blocking (bool) → Self

Move and/or cast 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.

Args:

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:

Module: self

Examples:

>>> # xdoctest: +IGNORE_WANT("non-deterministic")
>>> 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)
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA1)
>>> 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, recurse=True)

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

Args:

device (torch.device): The desired device of the parameters: and buffers in this module.
recurse (bool): Whether parameters and buffers of submodules should: be recursively moved to the specified device.

Returns:

Module: self

Parameters:

device (str | device | int | None)
recurse (bool)

Return type:

Self

train(mode=True)

Set the module in training mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e., whether they are affected, e.g. Dropout, BatchNorm, etc.

Args:

mode (bool): whether to set training mode (True) or evaluation: mode (False). Default: True.

Returns:

Module: self

Parameters:: mode (bool)
Return type:: Self

type(dst_type)

Casts all parameters and buffers to dst_type.

Note

This method modifies the module in-place.

Args:: dst_type (type or string): the desired type
Returns:: Module: self

Parameters:: dst_type (dtype | str)
Return type:: Self

xpu(device=None)

Move 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.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

zero_grad(set_to_none=True)

Reset gradients of all model parameters.

See similar function under torch.optim.Optimizer for more context.

Args:

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

Parameters:: set_to_none (bool)
Return type:: None

training: bool

King2019ModulationFilterbank

class torch_amt.King2019ModulationFilterbank(fs, mflow=2.0, mfhigh=150.0, qfactor=1.0, nmod=None, learnable=False, dtype=torch.float32)[source]

Bases: Module

Modulation filterbank for King et al. (2019) auditory model.

Implements the specific modulation filtering approach used in King2019, with logarithmically-spaced bandpass filters based on Q-factor and Butterworth 2nd-order design. This differs from the standard ModulationFilterbank which uses preset configurations (dau1997, jepsen2008).

The filterbank extracts amplitude modulation content from 2-150 Hz using bandpass filters with consistent Q-factor across all modulation frequencies.

Algorithm Overview

Center Frequency Spacing:

If nmod is None (automatic):

\[\begin{split}\\text{step} = \\frac{\\sqrt{4Q^2 + 1} + 1}{\\sqrt{4Q^2 + 1} - 1}\end{split}\]

\[\begin{split}\\log(f_{\\text{mod},i}) = \\log(f_{\\text{low}}) + i \\cdot \\log(\\text{step})\end{split}\]

If nmod is specified, use linear spacing in log domain.
Bandpass Limits:

For each center frequency \(f_c\):

\[\begin{split}f_{\\text{low}} = \\frac{f_c}{2}\\sqrt{4 + \\frac{1}{Q^2}} - \\frac{f_c}{2Q}\end{split}\]

\[\begin{split}f_{\\text{high}} = \\frac{f_c}{2}\\sqrt{4 + \\frac{1}{Q^2}} + \\frac{f_c}{2Q}\end{split}\]
Butterworth Bandpass:

2nd-order Butterworth for each modulation channel:

\[\begin{split}H_k(s) = \\frac{(\\omega_{\\text{high}} - \\omega_{\\text{low}})s}{s^2 + (\\omega_{\\text{high}} - \\omega_{\\text{low}})s + \\omega_{\\text{high}}\\omega_{\\text{low}}}\end{split}\]

param fs:: Sampling rate in Hz.
type fs:: float
param mflow:: Minimum modulation frequency in Hz. Default: 2.0.
type mflow:: float
param mfhigh:: Maximum modulation frequency in Hz. Default: 150.0.
type mfhigh:: float
param qfactor:: Q-factor for all modulation filters (bandwidth = mfc/Q). Default: 1.0.
type qfactor:: float
param nmod:: Number of modulation filters. If None, automatically determined by Q-factor spacing. Default: None (automatic).
type nmod:: int
param learnable:: If True, filter coefficients become trainable. Default: False.
type learnable:: bool
param dtype:: Data type for computations. Default: torch.float32.
type dtype:: dtype

mfc

Center frequencies of modulation filters, shape (n_filters,).

Type:: torch.Tensor

filters

List of SOSFilter modules, one per modulation channel.

Type:: nn.ModuleList

num_filters

Number of modulation filters.

Type:: int

Shape

-----

- Input

\(B\) = batch size
\(C\) = channels
\(T\) = time samples

Type:: \((B, C, T)\) where

\n - Output

\(M\) = number of modulation filters (num_filters)

Type:: \((B, C, M, T)\) where

Examples

>>> import torch
>>> from torch_amt.common.modulation import King2019ModulationFilterbank
>>>
>>> # Create filterbank
>>> mod_bank = King2019ModulationFilterbank(fs=48000, mflow=2.0, mfhigh=150.0, qfactor=1.0)
>>>
>>> # Input: (batch, channels, time)
>>> signal = torch.randn(2, 5, 2000)
>>>
>>> # Output: (batch, channels, n_mod_filters, time)
>>> output = mod_bank(signal)
>>> print(f"Input: {signal.shape}, Output: {output.shape}")
Input: torch.Size([2, 5, 2000]), Output: torch.Size([2, 5, 10, 2000])
>>>
>>> # Access center frequencies
>>> print(f"Modulation frequencies: {mod_bank.mfc}")
Modulation frequencies: tensor([  2.00,   4.83,  11.66, ..., 150.00])

Notes

Differences from ModulationFilterbank:

Uses logarithmic spacing based on Q-factor (not preset-based)
All filters are 2nd-order Butterworth bandpass
No lowpass filter at 0 Hz (starts at mflow)
No phase processing (preserves complex output if applicable)
Designed specifically for King2019 model

Numerical Stability:

Uses SOS (second-order sections) representation for all Butterworth filters, ensuring numerical stability even for extreme frequency ratios.

See also

ModulationFilterbank: Standard modulation filterbank with presets
ButterworthFilter: Butterworth filter design and application
SOSFilter: Second-order sections filtering

__init__(fs, mflow=2.0, mfhigh=150.0, qfactor=1.0, nmod=None, learnable=False, dtype=torch.float32)[source]

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

Parameters:

fs (float)
mflow (float)
mfhigh (float)
qfactor (float)
nmod (int)
learnable (bool)
dtype (dtype)

forward(x)[source]

Apply modulation filterbank.

Parameters:: x (Tensor) – Input signal, shape (batch, channels, time).
Returns:: Modulation-filtered output, shape (batch, channels, n_mod_filters, time).
Return type:: Tensor

extra_repr()[source]

Extra representation string for module printing.

Returns:: String containing key module parameters.
Return type:: str

T_destination = ~T_destination

add_module(name, module)

Add a child module to the current module.

The module can be accessed as an attribute using the given name.

Args:

name (str): name of the child module. The child module can be: accessed from this module using the given name

module (Module): child module to be added to the module.

Parameters:

name (str)
module (Module | None)

Return type:

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

Typical use includes initializing the parameters of a model (see also torch.nn.init).

Args:: fn (Module -> None): function to be applied to each submodule
Returns:: Module: self

Example:

>>> @torch.no_grad()
>>> def init_weights(m):
>>>     print(m)
>>>     if type(m) is nn.Linear:
>>>         m.weight.fill_(1.0)
>>>         print(m.weight)
>>> net = nn.Sequential(nn.Linear(2, 2), nn.Linear(2, 2))
>>> net.apply(init_weights)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)

Parameters:: fn (Callable[[Module], None])
Return type:: Self

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

buffers(recurse=True)

Return an iterator over module buffers.

Args:

recurse (bool): if True, then yields buffers of this module: and all submodules. Otherwise, yields only buffers that are direct members of this module.

Yields:

torch.Tensor: module buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for buf in model.buffers():
>>>     print(type(buf), buf.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Tensor]

call_super_init: bool = False

children()

Return an iterator over immediate children modules.

Return type:: Iterator[Module]

Yields:: Module: a child module

compile(*args, **kwargs)

Compile this Module’s forward using torch.compile().

This Module’s __call__ method is compiled and all arguments are passed as-is to torch.compile().

See torch.compile() for details on the arguments for this function.

Return type:: None

cpu()

Move all model parameters and buffers to the CPU.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

cuda(device=None)

Move all model parameters and buffers to the GPU.

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

Note

This method modifies the module in-place.

Args:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

double()

Casts all floating point parameters and buffers to double datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

dump_patches: bool = False

eval()

Set the module in evaluation mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e. whether 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.

Return type:: Self

Returns:: Module: self

float()

Casts all floating point parameters and buffers to float datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the buffer: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.Tensor: The buffer referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not a buffer

Parameters:: target (str)
Return type:: Tensor

get_extra_state()

Return 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 picklable to ensure working serialization of the state_dict. We only provide backwards compatibility guarantees for serializing Tensors; other objects may break backwards compatibility if their serialized pickled form changes.

Return type:: Any

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

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the Parameter: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.nn.Parameter: The Parameter referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not an nn.Parameter

Parameters:: target (str)
Return type:: Parameter

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(16, 33, kernel_size=(3, 3), stride=(2, 2))
        )
        (linear): Linear(in_features=100, out_features=200, bias=True)
    )
)

(The diagram shows an nn.Module A. A which 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.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

Returns:

torch.nn.Module: The submodule referenced by target

Raises:

AttributeError: If at any point along the path resulting from: the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:: target (str)
Return type:: Module

half()

Casts all floating point parameters and buffers to half datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

ipu(device=None)

Move all model parameters and buffers to the IPU.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

load_state_dict(state_dict, strict=True, assign=False)

Copy parameters and buffers from state_dict into this module and its descendants.

If strict is True, then the keys of state_dict must exactly match the keys returned by this module’s state_dict() function.

Warning

If assign is True the optimizer must be created after the call to load_state_dict unless get_swap_module_params_on_conversion() is True.

Args:

state_dict (dict): a dict containing parameters and: persistent buffers.
strict (bool, optional): whether to strictly enforce that the keys: in state_dict match the keys returned by this module’s state_dict() function. Default: True
assign (bool, optional): When set to False, the properties of the tensors: in the current module are preserved whereas setting it to True preserves properties of the Tensors in the state dict. The only exception is the requires_grad field of Parameter for which the value from the module is preserved. Default: False

Returns:

NamedTuple with missing_keys and unexpected_keys fields:

missing_keys is a list of str containing any keys that are expected
by this module but missing from the provided state_dict.
unexpected_keys is a list of str containing the keys that are not
expected by this module but present in the provided state_dict.

Note:

If a parameter or buffer is registered as None and its corresponding key exists in state_dict, load_state_dict() will raise a RuntimeError.

Parameters:

state_dict (Mapping[str, Any])
strict (bool)
assign (bool)

modules()

Return an iterator over all modules in the network.

Return type:: Iterator[Module]

Yields:: Module: a module in the network
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.modules()):
...     print(idx, '->', m)

0 -> Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)
1 -> Linear(in_features=2, out_features=2, bias=True)

mtia(device=None)

Move all model parameters and buffers to the MTIA.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

named_buffers(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

Args:

prefix (str): prefix to prepend to all buffer names. recurse (bool, optional): if True, then yields buffers of this module

and all submodules. Otherwise, yields only buffers that are direct members of this module. Defaults to True.

remove_duplicate (bool, optional): whether to remove the duplicated buffers in the result. Defaults to True.

Yields:

(str, torch.Tensor): Tuple containing the name and buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, buf in self.named_buffers():
>>>     if name in ['running_var']:
>>>         print(buf.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

Return type:: Iterator[tuple[str, Module]]

Yields:: (str, Module): Tuple containing a name and child module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, module in model.named_children():
>>>     if name in ['conv4', 'conv5']:
>>>         print(module)

named_modules(memo=None, prefix='', remove_duplicate=True)

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

Args:: memo: a memo to store the set of modules already added to the result prefix: a prefix that will be added to the name of the module remove_duplicate: whether to remove the duplicated module instances in the result

or not
Yields:: (str, Module): Tuple of name and module
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.named_modules()):
...     print(idx, '->', m)

0 -> ('', Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
))
1 -> ('0', Linear(in_features=2, out_features=2, bias=True))

Parameters:

memo (set[Module] | None)
prefix (str)
remove_duplicate (bool)

named_parameters(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

Args:

prefix (str): prefix to prepend to all parameter names. recurse (bool): if True, then yields parameters of this module

and all submodules. Otherwise, yields only parameters that are direct members of this module.

remove_duplicate (bool, optional): whether to remove the duplicated: parameters in the result. Defaults to True.

Yields:

(str, Parameter): Tuple containing the name and parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, param in self.named_parameters():
>>>     if name in ['bias']:
>>>         print(param.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

parameters(recurse=True)

Return an iterator over module parameters.

This is typically passed to an optimizer.

Args:

recurse (bool): if True, then yields parameters of this module: and all submodules. Otherwise, yields only parameters that are direct members of this module.

Yields:

Parameter: module parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for param in model.parameters():
>>>     print(type(param), param.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Parameter]

register_backward_hook(hook)

Register 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:

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

Parameters:: hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
Return type:: RemovableHandle

register_buffer(name, tensor, persistent=True)

Add a buffer to the module.

This is typically used to register a buffer that should not be considered a model parameter. For example, BatchNorm’s running_mean is not a parameter, but is part of the module’s state. Buffers, by default, are persistent and will be saved alongside parameters. This behavior can be changed by setting persistent to False. The only difference between a persistent buffer and a non-persistent buffer is that the latter will not be a part of this module’s state_dict.

Buffers can be accessed as attributes using given names.

Args:

name (str): name of the buffer. The buffer can be accessed: from this module using the given name
tensor (Tensor or None): buffer to be registered. If None, then operations: that run on buffers, such as cuda, are ignored. If None, the buffer is not included in the module’s state_dict.
persistent (bool): whether the buffer is part of this module’s: state_dict.

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> self.register_buffer('running_mean', torch.zeros(num_features))

Parameters:

name (str)
tensor (Tensor | None)
persistent (bool)

Return type:

register_forward_hook(hook, *, prepend=False, with_kwargs=False, always_call=False)

Register a forward hook on the module.

The hook will be called every time after forward() has computed an output.

If with_kwargs is False or not specified, 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. The hook should have the following signature:

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

If with_kwargs is True, the forward hook will be passed the kwargs given to the forward function and be expected to return the output possibly modified. The hook should have the following signature:

hook(module, args, kwargs, output) -> None or modified output

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If True, the provided hook will be fired

before all existing forward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward hooks on this torch.nn.Module. Note that global forward hooks registered with register_module_forward_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If True, the hook will be passed the: kwargs given to the forward function. Default: False
always_call (bool): If True the hook will be run regardless of: whether an exception is raised while calling the Module. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...], Any], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any], Any], Any | None])
prepend (bool)
with_kwargs (bool)
always_call (bool)

Return type:

RemovableHandle

register_forward_pre_hook(hook, *, prepend=False, with_kwargs=False)

Register a forward pre-hook on the module.

The hook will be called every time before forward() is invoked.

If with_kwargs is false or not specified, 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). The hook should have the following signature:

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

If with_kwargs is true, the forward pre-hook will be passed the kwargs given to the forward function. And if the hook modifies the input, both the args and kwargs should be returned. The hook should have the following signature:

hook(module, args, kwargs) -> None or a tuple of modified input and kwargs

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing forward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward_pre hooks on this torch.nn.Module. Note that global forward_pre hooks registered with register_module_forward_pre_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If true, the hook will be passed the kwargs: given to the forward function. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...]], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any]], tuple[Any, dict[str, Any]] | None])
prepend (bool)
with_kwargs (bool)

Return type:

RemovableHandle

register_full_backward_hook(hook, prepend=False)

Register a backward hook on the module.

The hook will be called every time the gradients with respect to a module are computed, and its firing rules are as follows:

Ordinarily, the hook fires when the gradients are computed with respect to the module inputs.

If none of the module inputs require gradients, the hook will fire when the gradients are computed with respect to module outputs.

If none of the module outputs require gradients, then the hooks will not fire.

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.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward hooks on this torch.nn.Module. Note that global backward hooks registered with register_module_full_backward_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_full_backward_pre_hook(hook, prepend=False)

Register a backward pre-hook on the module.

The hook will be called every time the gradients for the module are computed. The hook should have the following signature:

hook(module, grad_output) -> tuple[Tensor, ...], Tensor or None

The grad_output is a tuple. The hook should not modify its arguments, but it can optionally return a new gradient with respect to the output that will be used in place of grad_output in subsequent computations. Entries in 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 inplace is not allowed when using backward hooks and will raise an error.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward_pre hooks on this torch.nn.Module. Note that global backward_pre hooks registered with register_module_full_backward_pre_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module’s load_state_dict() is called.

It should have the following signature::: hook(module, incompatible_keys) -> None

The module argument is the current module that this hook is registered on, and the incompatible_keys argument is a NamedTuple consisting of attributes missing_keys and unexpected_keys. missing_keys is a list of str containing the missing keys and unexpected_keys is a list of str containing the unexpected keys.

The given incompatible_keys can be modified inplace if needed.

Note that the checks performed when calling load_state_dict() with strict=True are affected by modifications the hook makes to missing_keys or unexpected_keys, as expected. Additions to either set of keys will result in an error being thrown when strict=True, and clearing out both missing and unexpected keys will avoid an error.

Returns:

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

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module’s load_state_dict() is called.

It should have the following signature::

hook(module, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) -> None # noqa: B950

Arguments:

hook (Callable): Callable hook that will be invoked before: loading the state dict.

register_module(name, module)

Alias for add_module().

Parameters:

name (str)
module (Module | None)

Return type:

register_parameter(name, param)

Add a parameter to the module.

The parameter can be accessed as an attribute using given name.

Args:

name (str): name of the parameter. The parameter can be accessed: from this module using the given name
param (Parameter or None): parameter to be added to the module. If: None, then operations that run on parameters, such as cuda, are ignored. If None, the parameter is not included in the module’s state_dict.

Parameters:

name (str)
param (Parameter | None)

Return type:

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

It should have the following signature::: hook(module, state_dict, prefix, local_metadata) -> None

The registered hooks can modify the state_dict inplace.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

It should have the following signature::: hook(module, prefix, keep_vars) -> None

The registered hooks can be used to perform pre-processing before the state_dict call is made.

requires_grad_(requires_grad=True)

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.

Args:

requires_grad (bool): whether autograd should record operations on: parameters in this module. Default: True.

Returns:

Module: self

Parameters:: requires_grad (bool)
Return type:: Self

set_extra_state(state)

Set extra state contained in the loaded state_dict.

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.

Args:: state (dict): Extra state from the state_dict

Parameters:: state (Any)
Return type:: None

set_submodule(target, module, strict=False)

Set the submodule given by target if it exists, otherwise throw an error.

Note

If strict is set to False (default), the method will replace an existing submodule or create a new submodule if the parent module exists. If strict is set to True, the method will only attempt to replace an existing submodule and throw an error if the submodule does not exist.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(3, 3, 3)
        )
        (linear): Linear(3, 3)
    )
)

(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 override the Conv2d with a new submodule Linear, you could call set_submodule("net_b.net_c.conv", nn.Linear(1, 1)) where strict could be True or False

To add a new submodule Conv2d to the existing net_b module, you would call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1)).

In the above if you set strict=True and call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1), strict=True), an AttributeError will be raised because net_b does not have a submodule named conv.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

module: The module to set the submodule to. strict: If False, the method will replace an existing submodule

or create a new submodule if the parent module exists. If True, the method will only attempt to replace an existing submodule and throw an error if the submodule doesn’t already exist.

Raises:

ValueError: If the target string is empty or if module is not an instance of nn.Module. AttributeError: If at any point along the path resulting from

the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:

target (str)
module (Module)
strict (bool)

Return type:

share_memory()

Return type:: Self

state_dict(*args, destination=None, prefix='', keep_vars=False)

Overloads:

self, destination (T_destination), prefix (str), keep_vars (bool) → T_destination
self, prefix (str), keep_vars (bool) → dict[str, Any]

Return a dictionary containing references to the whole state of the module.

Both parameters and persistent buffers (e.g. running averages) are included. Keys are corresponding parameter and buffer names. Parameters and buffers set to None are not included.

Note

The returned object is a shallow copy. It contains references to the module’s parameters and buffers.

Warning

Currently state_dict() also accepts positional arguments for destination, prefix and keep_vars in order. However, this is being deprecated and keyword arguments will be enforced in future releases.

Warning

Please avoid the use of argument destination as it is not designed for end-users.

Args:

destination (dict, optional): If provided, the state of module will: be updated into the dict and the same object is returned. Otherwise, an OrderedDict will be created and returned. Default: None.
prefix (str, optional): a prefix added to parameter and buffer: names to compose the keys in state_dict. Default: ''.
keep_vars (bool, optional): by default the Tensor s: returned in the state dict are detached from autograd. If it’s set to True, detaching will not be performed. Default: False.

Returns:

dict:: a dictionary containing a whole state of the module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> module.state_dict().keys()
['bias', 'weight']

to(*args, **kwargs)

Overloads:

self, device (DeviceLikeType | None), dtype (dtype | None), non_blocking (bool) → Self
self, dtype (dtype), non_blocking (bool) → Self
self, tensor (Tensor), non_blocking (bool) → Self

Move and/or cast 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.

Args:

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:

Module: self

Examples:

>>> # xdoctest: +IGNORE_WANT("non-deterministic")
>>> 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)
>>> # xdoctest: +REQUIRES(env:TORCH_DOCTEST_CUDA1)
>>> 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, recurse=True)

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

Args:

device (torch.device): The desired device of the parameters: and buffers in this module.
recurse (bool): Whether parameters and buffers of submodules should: be recursively moved to the specified device.

Returns:

Module: self

Parameters:

device (str | device | int | None)
recurse (bool)

Return type:

Self

train(mode=True)

Set the module in training mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e., whether they are affected, e.g. Dropout, BatchNorm, etc.

Args:

mode (bool): whether to set training mode (True) or evaluation: mode (False). Default: True.

Returns:

Module: self

Parameters:: mode (bool)
Return type:: Self

type(dst_type)

Casts all parameters and buffers to dst_type.

Note

This method modifies the module in-place.

Args:: dst_type (type or string): the desired type
Returns:: Module: self

Parameters:: dst_type (dtype | str)
Return type:: Self

xpu(device=None)

Move 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.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

zero_grad(set_to_none=True)

Reset gradients of all model parameters.

See similar function under torch.optim.Optimizer for more context.

Args:

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

Parameters:: set_to_none (bool)
Return type:: None

training: bool

FastKing2019ModulationFilterbank

class torch_amt.FastKing2019ModulationFilterbank(fs, mflow=2.0, mfhigh=150.0, qfactor=1.0, nmod=None, learnable=False, dtype=torch.float32, ir_length_ratio=1.0)[source]

Bases: Module

Fast FFT-based modulation filterbank for King et al. (2019) model.

Drop-in replacement for King2019ModulationFilterbank that uses FFT-based convolution instead of recursive IIR filtering. Achieves ~250x speedup with ~15% output difference and ~13% gradient difference.

Trade-offs:

Speed: ~250x faster (2.6s → 0.01s for 0.5s @ 48kHz)
Accuracy: ~15% relative output error, ~13% gradient error
Use case: Ideal for inference/feature extraction, acceptable for training when approximate gradients are sufficient

Implementation:

Instead of applying each filter recursively using IIR structure, this implementation:

Pre-computes impulse responses for all filters
Applies FFT to input and impulse responses
Multiplies in frequency domain (parallel!)
Applies IFFT to get output

This is mathematically equivalent to linear convolution, which approximates the infinite impulse response with a finite-length impulse response.

Parameters:

fs (float) – Sampling rate in Hz.
mflow (float) – Minimum modulation frequency in Hz. Default: 2.0.
mfhigh (float) – Maximum modulation frequency in Hz. Default: 150.0.
qfactor (float) – Q-factor for all modulation filters. Default: 1.0.
nmod (int) – Number of modulation filters. If None, automatically determined. Default: None (automatic).
learnable (bool) – If True, filter coefficients become trainable. Default: False. Note: Learnable mode uses the same FFT convolution, so gradients will have ~13% error compared to exact IIR backprop.
dtype (dtype) – Data type for computations. Default: torch.float32.
ir_length_ratio (float) – Ratio of signal length to use for impulse response. Default: 1.0 (use full signal length for maximum accuracy). Can be reduced (e.g., 0.5) for faster computation with slightly more error.

mfc

Center frequencies of modulation filters, shape (n_filters,).

Type:: torch.Tensor

sos_stack

Stacked SOS coefficients for all filters, shape (n_filters, n_sections, 6).

Type:: torch.Tensor

num_filters

Number of modulation filters.

Type:: int

ir_length_ratio

Ratio of signal length used for impulse response computation.

Type:: float

Shape

-----

- Input

\(B\) = batch size
\(C\) = channels
\(T\) = time samples

Type:: \((B, C, T)\) where

- Output

\(M\) = number of modulation filters (num_filters)

Type:: \((B, C, M, T)\) where

Examples

>>> import torch
>>> from torch_amt.common.modulation import FastKing2019ModulationFilterbank
>>>
>>> # Create filterbank
>>> mod_bank = FastKing2019ModulationFilterbank(fs=48000, mflow=2.0, mfhigh=150.0)
>>>
>>> # Input: (batch, channels, time)
>>> signal = torch.randn(2, 5, 24000)
>>>
>>> # Output: (batch, channels, n_mod_filters, time)
>>> output = mod_bank(signal)
>>> print(f"Shape: {output.shape}")
Shape: torch.Size([2, 5, 5, 24000])

Notes

Accuracy vs Speed:

The FFT-based approach introduces small numerical differences compared to the exact IIR recursive implementation:

Output: ~15% relative error (1-3% absolute for normalized signals)
Gradients: ~13% relative error
Speedup: ~250x faster

These differences arise from:

Truncating infinite impulse response to finite length
Different numerical precision in FFT vs recursive computation
No recursive error accumulation (FFT is more stable numerically)

Backward compatibility:

This is a drop-in replacement for King2019ModulationFilterbank with identical API. Simply replace the class name to enable fast mode.

See also

King2019ModulationFilterbank: Exact IIR implementation (slow but precise)
FastModulationFilterbank: FFT-based implementation for Dau-style filterbanks

References

__init__(fs, mflow=2.0, mfhigh=150.0, qfactor=1.0, nmod=None, learnable=False, dtype=torch.float32, ir_length_ratio=1.0)[source]

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

Parameters:

fs (float)
mflow (float)
mfhigh (float)
qfactor (float)
nmod (int)
learnable (bool)
dtype (dtype)
ir_length_ratio (float)

forward(x)[source]

Apply modulation filterbank using FFT-based convolution.

All operations are PyTorch-native to preserve gradient flow!

Parameters:: x (Tensor) – Input signal, shape (batch, channels, time).
Returns:: Modulation-filtered output, shape (batch, channels, n_mod_filters, time).
Return type:: Tensor

T_destination = ~T_destination

add_module(name, module)

Add a child module to the current module.

The module can be accessed as an attribute using the given name.

Args:

name (str): name of the child module. The child module can be: accessed from this module using the given name

module (Module): child module to be added to the module.

Parameters:

name (str)
module (Module | None)

Return type:

apply(fn)

Apply fn recursively to every submodule (as returned by .children()) as well as self.

Typical use includes initializing the parameters of a model (see also torch.nn.init).

Args:: fn (Module -> None): function to be applied to each submodule
Returns:: Module: self

Example:

>>> @torch.no_grad()
>>> def init_weights(m):
>>>     print(m)
>>>     if type(m) is nn.Linear:
>>>         m.weight.fill_(1.0)
>>>         print(m.weight)
>>> net = nn.Sequential(nn.Linear(2, 2), nn.Linear(2, 2))
>>> net.apply(init_weights)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Linear(in_features=2, out_features=2, bias=True)
Parameter containing:
tensor([[1., 1.],
        [1., 1.]], requires_grad=True)
Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)

Parameters:: fn (Callable[[Module], None])
Return type:: Self

bfloat16()

Casts all floating point parameters and buffers to bfloat16 datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

buffers(recurse=True)

Return an iterator over module buffers.

Args:

recurse (bool): if True, then yields buffers of this module: and all submodules. Otherwise, yields only buffers that are direct members of this module.

Yields:

torch.Tensor: module buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for buf in model.buffers():
>>>     print(type(buf), buf.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Tensor]

call_super_init: bool = False

children()

Return an iterator over immediate children modules.

Return type:: Iterator[Module]

Yields:: Module: a child module

compile(*args, **kwargs)

Compile this Module’s forward using torch.compile().

This Module’s __call__ method is compiled and all arguments are passed as-is to torch.compile().

See torch.compile() for details on the arguments for this function.

Return type:: None

cpu()

Move all model parameters and buffers to the CPU.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

cuda(device=None)

Move all model parameters and buffers to the GPU.

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

Note

This method modifies the module in-place.

Args:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

double()

Casts all floating point parameters and buffers to double datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

dump_patches: bool = False

eval()

Set the module in evaluation mode.

This has an effect only on certain modules. See the documentation of particular modules for details of their behaviors in training/evaluation mode, i.e. whether 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.

Return type:: Self

Returns:: Module: self

extra_repr()[source]

Extra representation string for module printing.

Returns:: String containing key module parameters.
Return type:: str

float()

Casts all floating point parameters and buffers to float datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

get_buffer(target)

Return the buffer given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the buffer: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.Tensor: The buffer referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not a buffer

Parameters:: target (str)
Return type:: Tensor

get_extra_state()

Return 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 picklable to ensure working serialization of the state_dict. We only provide backwards compatibility guarantees for serializing Tensors; other objects may break backwards compatibility if their serialized pickled form changes.

Return type:: Any

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

get_parameter(target)

Return the parameter given by target if it exists, otherwise throw an error.

See the docstring for get_submodule for a more detailed explanation of this method’s functionality as well as how to correctly specify target.

Args:

target: The fully-qualified string name of the Parameter: to look for. (See get_submodule for how to specify a fully-qualified string.)

Returns:

torch.nn.Parameter: The Parameter referenced by target

Raises:

AttributeError: If the target string references an invalid: path or resolves to something that is not an nn.Parameter

Parameters:: target (str)
Return type:: Parameter

get_submodule(target)

Return the submodule given by target if it exists, otherwise throw an error.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(16, 33, kernel_size=(3, 3), stride=(2, 2))
        )
        (linear): Linear(in_features=100, out_features=200, bias=True)
    )
)

(The diagram shows an nn.Module A. A which 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.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

Returns:

torch.nn.Module: The submodule referenced by target

Raises:

AttributeError: If at any point along the path resulting from: the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:: target (str)
Return type:: Module

half()

Casts all floating point parameters and buffers to half datatype.

Note

This method modifies the module in-place.

Returns:: Module: self

Return type:: Self

ipu(device=None)

Move all model parameters and buffers to the IPU.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

load_state_dict(state_dict, strict=True, assign=False)

Copy parameters and buffers from state_dict into this module and its descendants.

If strict is True, then the keys of state_dict must exactly match the keys returned by this module’s state_dict() function.

Warning

If assign is True the optimizer must be created after the call to load_state_dict unless get_swap_module_params_on_conversion() is True.

Args:

state_dict (dict): a dict containing parameters and: persistent buffers.
strict (bool, optional): whether to strictly enforce that the keys: in state_dict match the keys returned by this module’s state_dict() function. Default: True
assign (bool, optional): When set to False, the properties of the tensors: in the current module are preserved whereas setting it to True preserves properties of the Tensors in the state dict. The only exception is the requires_grad field of Parameter for which the value from the module is preserved. Default: False

Returns:

NamedTuple with missing_keys and unexpected_keys fields:

missing_keys is a list of str containing any keys that are expected
by this module but missing from the provided state_dict.
unexpected_keys is a list of str containing the keys that are not
expected by this module but present in the provided state_dict.

Note:

If a parameter or buffer is registered as None and its corresponding key exists in state_dict, load_state_dict() will raise a RuntimeError.

Parameters:

state_dict (Mapping[str, Any])
strict (bool)
assign (bool)

modules()

Return an iterator over all modules in the network.

Return type:: Iterator[Module]

Yields:: Module: a module in the network
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.modules()):
...     print(idx, '->', m)

0 -> Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
)
1 -> Linear(in_features=2, out_features=2, bias=True)

mtia(device=None)

Move all model parameters and buffers to the MTIA.

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

Note

This method modifies the module in-place.

Arguments:

device (int, optional): if specified, all parameters will be: copied to that device

Returns:

Module: self

Parameters:: device (int | device | None)
Return type:: Self

named_buffers(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.

Args:

prefix (str): prefix to prepend to all buffer names. recurse (bool, optional): if True, then yields buffers of this module

and all submodules. Otherwise, yields only buffers that are direct members of this module. Defaults to True.

remove_duplicate (bool, optional): whether to remove the duplicated buffers in the result. Defaults to True.

Yields:

(str, torch.Tensor): Tuple containing the name and buffer

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, buf in self.named_buffers():
>>>     if name in ['running_var']:
>>>         print(buf.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

named_children()

Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.

Return type:: Iterator[tuple[str, Module]]

Yields:: (str, Module): Tuple containing a name and child module

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, module in model.named_children():
>>>     if name in ['conv4', 'conv5']:
>>>         print(module)

named_modules(memo=None, prefix='', remove_duplicate=True)

Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.

Args:: memo: a memo to store the set of modules already added to the result prefix: a prefix that will be added to the name of the module remove_duplicate: whether to remove the duplicated module instances in the result

or not
Yields:: (str, Module): Tuple of name and module
Note:: Duplicate modules are returned only once. In the following example, l will be returned only once.

Example:

>>> l = nn.Linear(2, 2)
>>> net = nn.Sequential(l, l)
>>> for idx, m in enumerate(net.named_modules()):
...     print(idx, '->', m)

0 -> ('', Sequential(
  (0): Linear(in_features=2, out_features=2, bias=True)
  (1): Linear(in_features=2, out_features=2, bias=True)
))
1 -> ('0', Linear(in_features=2, out_features=2, bias=True))

Parameters:

memo (set[Module] | None)
prefix (str)
remove_duplicate (bool)

named_parameters(prefix='', recurse=True, remove_duplicate=True)

Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.

Args:

prefix (str): prefix to prepend to all parameter names. recurse (bool): if True, then yields parameters of this module

and all submodules. Otherwise, yields only parameters that are direct members of this module.

remove_duplicate (bool, optional): whether to remove the duplicated: parameters in the result. Defaults to True.

Yields:

(str, Parameter): Tuple containing the name and parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for name, param in self.named_parameters():
>>>     if name in ['bias']:
>>>         print(param.size())

Parameters:

prefix (str)
recurse (bool)
remove_duplicate (bool)

Return type:

parameters(recurse=True)

Return an iterator over module parameters.

This is typically passed to an optimizer.

Args:

recurse (bool): if True, then yields parameters of this module: and all submodules. Otherwise, yields only parameters that are direct members of this module.

Yields:

Parameter: module parameter

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> for param in model.parameters():
>>>     print(type(param), param.size())
<class 'torch.Tensor'> (20L,)
<class 'torch.Tensor'> (20L, 1L, 5L, 5L)

Parameters:: recurse (bool)
Return type:: Iterator[Parameter]

register_backward_hook(hook)

Register 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:

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

Parameters:: hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
Return type:: RemovableHandle

register_buffer(name, tensor, persistent=True)

Add a buffer to the module.

This is typically used to register a buffer that should not be considered a model parameter. For example, BatchNorm’s running_mean is not a parameter, but is part of the module’s state. Buffers, by default, are persistent and will be saved alongside parameters. This behavior can be changed by setting persistent to False. The only difference between a persistent buffer and a non-persistent buffer is that the latter will not be a part of this module’s state_dict.

Buffers can be accessed as attributes using given names.

Args:

name (str): name of the buffer. The buffer can be accessed: from this module using the given name
tensor (Tensor or None): buffer to be registered. If None, then operations: that run on buffers, such as cuda, are ignored. If None, the buffer is not included in the module’s state_dict.
persistent (bool): whether the buffer is part of this module’s: state_dict.

Example:

>>> # xdoctest: +SKIP("undefined vars")
>>> self.register_buffer('running_mean', torch.zeros(num_features))

Parameters:

name (str)
tensor (Tensor | None)
persistent (bool)

Return type:

register_forward_hook(hook, *, prepend=False, with_kwargs=False, always_call=False)

Register a forward hook on the module.

The hook will be called every time after forward() has computed an output.

If with_kwargs is False or not specified, 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. The hook should have the following signature:

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

If with_kwargs is True, the forward hook will be passed the kwargs given to the forward function and be expected to return the output possibly modified. The hook should have the following signature:

hook(module, args, kwargs, output) -> None or modified output

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If True, the provided hook will be fired

before all existing forward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward hooks on this torch.nn.Module. Note that global forward hooks registered with register_module_forward_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If True, the hook will be passed the: kwargs given to the forward function. Default: False
always_call (bool): If True the hook will be run regardless of: whether an exception is raised while calling the Module. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...], Any], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any], Any], Any | None])
prepend (bool)
with_kwargs (bool)
always_call (bool)

Return type:

RemovableHandle

register_forward_pre_hook(hook, *, prepend=False, with_kwargs=False)

Register a forward pre-hook on the module.

The hook will be called every time before forward() is invoked.

If with_kwargs is false or not specified, 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). The hook should have the following signature:

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

If with_kwargs is true, the forward pre-hook will be passed the kwargs given to the forward function. And if the hook modifies the input, both the args and kwargs should be returned. The hook should have the following signature:

hook(module, args, kwargs) -> None or a tuple of modified input and kwargs

Args:

hook (Callable): The user defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing forward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing forward_pre hooks on this torch.nn.Module. Note that global forward_pre hooks registered with register_module_forward_pre_hook() will fire before all hooks registered by this method. Default: False

with_kwargs (bool): If true, the hook will be passed the kwargs: given to the forward function. Default: False

Returns:

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

Parameters:

hook (Callable[[TypeVar(T, bound= Module), tuple[Any, ...]], Any | None] | Callable[[TypeVar(T, bound= Module), tuple[Any, ...], dict[str, Any]], tuple[Any, dict[str, Any]] | None])
prepend (bool)
with_kwargs (bool)

Return type:

RemovableHandle

register_full_backward_hook(hook, prepend=False)

Register a backward hook on the module.

The hook will be called every time the gradients with respect to a module are computed, and its firing rules are as follows:

Ordinarily, the hook fires when the gradients are computed with respect to the module inputs.

If none of the module inputs require gradients, the hook will fire when the gradients are computed with respect to module outputs.

If none of the module outputs require gradients, then the hooks will not fire.

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.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward hooks on this torch.nn.Module. Note that global backward hooks registered with register_module_full_backward_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_full_backward_pre_hook(hook, prepend=False)

Register a backward pre-hook on the module.

The hook will be called every time the gradients for the module are computed. The hook should have the following signature:

hook(module, grad_output) -> tuple[Tensor, ...], Tensor or None

The grad_output is a tuple. The hook should not modify its arguments, but it can optionally return a new gradient with respect to the output that will be used in place of grad_output in subsequent computations. Entries in 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 inplace is not allowed when using backward hooks and will raise an error.

Args:

hook (Callable): The user-defined hook to be registered. prepend (bool): If true, the provided hook will be fired before

all existing backward_pre hooks on this torch.nn.Module. Otherwise, the provided hook will be fired after all existing backward_pre hooks on this torch.nn.Module. Note that global backward_pre hooks registered with register_module_full_backward_pre_hook() will fire before all hooks registered by this method.

Returns:

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

Parameters:

hook (Callable[[Module, tuple[Tensor, ...] | Tensor], None | tuple[Tensor, ...] | Tensor])
prepend (bool)

Return type:

RemovableHandle

register_load_state_dict_post_hook(hook)

Register a post-hook to be run after module’s load_state_dict() is called.

It should have the following signature::: hook(module, incompatible_keys) -> None

The module argument is the current module that this hook is registered on, and the incompatible_keys argument is a NamedTuple consisting of attributes missing_keys and unexpected_keys. missing_keys is a list of str containing the missing keys and unexpected_keys is a list of str containing the unexpected keys.

The given incompatible_keys can be modified inplace if needed.

Note that the checks performed when calling load_state_dict() with strict=True are affected by modifications the hook makes to missing_keys or unexpected_keys, as expected. Additions to either set of keys will result in an error being thrown when strict=True, and clearing out both missing and unexpected keys will avoid an error.

Returns:

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

register_load_state_dict_pre_hook(hook)

Register a pre-hook to be run before module’s load_state_dict() is called.

It should have the following signature::

hook(module, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs) -> None # noqa: B950

Arguments:

hook (Callable): Callable hook that will be invoked before: loading the state dict.

register_module(name, module)

Alias for add_module().

Parameters:

name (str)
module (Module | None)

Return type:

register_parameter(name, param)

Add a parameter to the module.

The parameter can be accessed as an attribute using given name.

Args:

name (str): name of the parameter. The parameter can be accessed: from this module using the given name
param (Parameter or None): parameter to be added to the module. If: None, then operations that run on parameters, such as cuda, are ignored. If None, the parameter is not included in the module’s state_dict.

Parameters:

name (str)
param (Parameter | None)

Return type:

register_state_dict_post_hook(hook)

Register a post-hook for the state_dict() method.

It should have the following signature::: hook(module, state_dict, prefix, local_metadata) -> None

The registered hooks can modify the state_dict inplace.

register_state_dict_pre_hook(hook)

Register a pre-hook for the state_dict() method.

It should have the following signature::: hook(module, prefix, keep_vars) -> None

The registered hooks can be used to perform pre-processing before the state_dict call is made.

requires_grad_(requires_grad=True)

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.

Args:

requires_grad (bool): whether autograd should record operations on: parameters in this module. Default: True.

Returns:

Module: self

Parameters:: requires_grad (bool)
Return type:: Self

set_extra_state(state)

Set extra state contained in the loaded state_dict.

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.

Args:: state (dict): Extra state from the state_dict

Parameters:: state (Any)
Return type:: None

set_submodule(target, module, strict=False)

Set the submodule given by target if it exists, otherwise throw an error.

Note

If strict is set to False (default), the method will replace an existing submodule or create a new submodule if the parent module exists. If strict is set to True, the method will only attempt to replace an existing submodule and throw an error if the submodule does not exist.

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

A(
    (net_b): Module(
        (net_c): Module(
            (conv): Conv2d(3, 3, 3)
        )
        (linear): Linear(3, 3)
    )
)

(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 override the Conv2d with a new submodule Linear, you could call set_submodule("net_b.net_c.conv", nn.Linear(1, 1)) where strict could be True or False

To add a new submodule Conv2d to the existing net_b module, you would call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1)).

In the above if you set strict=True and call set_submodule("net_b.conv", nn.Conv2d(1, 1, 1), strict=True), an AttributeError will be raised because net_b does not have a submodule named conv.

Args:

target: The fully-qualified string name of the submodule: to look for. (See above example for how to specify a fully-qualified string.)

module: The module to set the submodule to. strict: If False, the method will replace an existing submodule

or create a new submodule if the parent module exists. If True, the method will only attempt to replace an existing submodule and throw an error if the submodule doesn’t already exist.

Raises:

ValueError: If the target string is empty or if module is not an instance of nn.Module. AttributeError: If at any point along the path resulting from

the target string the (sub)path resolves to a non-existent attribute name or an object that is not an instance of nn.Module.

Parameters:

target (str)
module (Module)
strict (bool)

Return type:

share_memory()