aihwkit.simulator.tiles.inference_torch module

High level analog tiles (inference).

class aihwkit.simulator.tiles.inference_torch.InputRangeForward(*args, **kwargs)

Bases: Function

Enable custom input range gradient computation using torch’s autograd.

static backward(ctx, d_output)

Define a formula for differentiating the operation with backward mode automatic differentiation.

This function is to be overridden by all subclasses. (Defining this function is equivalent to defining the vjp function.)

It must accept a context ctx as the first argument, followed by as many outputs as the forward() returned (None will be passed in for non tensor outputs of the forward function), and it should return as many tensors, as there were inputs to forward(). Each argument is the gradient w.r.t the given output, and each returned value should be the gradient w.r.t. the corresponding input. If an input is not a Tensor or is a Tensor not requiring grads, you can just pass None as a gradient for that input.

The context can be used to retrieve tensors saved during the forward pass. It also has an attribute ctx.needs_input_grad as a tuple of booleans representing whether each input needs gradient. E.g., backward() will have ctx.needs_input_grad[0] = True if the first input to forward() needs gradient computed w.r.t. the output.

Parameters:

• ctx (Any) –

• d_output (Tensor) –

Return type:

Tuple[Tensor, Tensor, None]

static forward(ctx, x_input, input_range, ir_params)

Define the forward of the custom autograd Function.

This function is to be overridden by all subclasses. There are two ways to define forward:

Usage 1 (Combined forward and ctx):

@staticmethod
def forward(ctx: Any, *args: Any, **kwargs: Any) -> Any:
    pass

• It must accept a context ctx as the first argument, followed by any number of arguments (tensors or other types).

• See combining-forward-context for more details

Usage 2 (Separate forward and ctx):

@staticmethod
def forward(*args: Any, **kwargs: Any) -> Any:
    pass

@staticmethod
def setup_context(ctx: Any, inputs: Tuple[Any, ...], output: Any) -> None:
    pass

• The forward no longer accepts a ctx argument.

• Instead, you must also override the torch.autograd.Function.setup_context() staticmethod to handle setting up the ctx object. output is the output of the forward, inputs are a Tuple of inputs to the forward.

• See extending-autograd for more details

The context can be used to store arbitrary data that can be then retrieved during the backward pass. Tensors should not be stored directly on ctx (though this is not currently enforced for backward compatibility). Instead, tensors should be saved either with ctx.save_for_backward() if they are intended to be used in backward (equivalently, vjp) or ctx.save_for_forward() if they are intended to be used for in jvp.

Parameters:

• ctx (Any) –

• x_input (Tensor) –

• input_range (Tensor) –

• ir_params (InputRangeParameter) –

Return type:

Tensor

class aihwkit.simulator.tiles.inference_torch.TorchInferenceTile(out_size, in_size, rpu_config=None, bias=False, in_trans=False, out_trans=False)

Bases: TileModule, InferenceTileWithPeriphery, SimulatorTileWrapper

InferenceTile using a torch-based simulator tile (and not a tile from RPUCuda).

Parameters:

• out_size (int) –

• in_size (int) –

• rpu_config (TorchInferenceRPUConfig | None) –

• bias (bool) –

• in_trans (bool) –

• out_trans (bool) –

forward(x_input, tensor_view=None)

Torch forward function that calls the analog forward

Parameters:

• x_input (Tensor) –

• tensor_view (Tuple | None) –

Return type:

Tensor

get_forward_parameters()

Get the additional parameters generated for the forward pass.

Returns:

Dictionary of the forward parameters set.

Return type:

Dict[str, Tensor]

init_input_processing()

Helper function to initialize the input processing.

Note

This method is called from the constructor.

Returns:

whether input processing is enabled

Return type:

bool

Raises: ConfigError in case manage_output_clipping is

enabled but not supported.

post_update_step()

Clip and remap weights after weights have been updated.

Return type:

None

pre_forward(x_input, dim, is_test=False, ctx=None)

Operations before the actual forward step for pre processing.

By default, this is an no-op. However, it could be overridden in derived tile classes.

Parameters:

• x_input (Tensor) – input tensor for the analog MVM of the tile.

• dim (int) – input channel dimension, ie the x_size dimension

• is_test (bool) – whether in eval mode

• ctx (Any | None) – torch auto-grad context [Optional]

Returns:

Output tensor of the same shape

Return type:

Tensor

set_forward_parameters(dic=None, **kwargs)

Set the additional parameters generated for the forward pass.

Currently only out_noise_values is implemented.

Parameters:

• dic (Dict[str, Tensor] | None) – dictionary of parameters to set (from get_forward_parameter())

• kwargs (Dict[str, Tensor]) – parameter names can alternatively given directly as keywords

Raises:

ArgumentError – If size are mismatched or keyword unknown

Return type:

None

set_scales(scales)

Set all scales with a new scale.

This will set the mapping scales to scales and set all other scales to 1.

Parameters:

scales (Tensor) – scales to set.

Return type:

None

supports_ddp: bool = True

supports_indexed: bool = False