ml.models.architectures.unet

Defines a general-purpose UNet model.

class ml.models.architectures.unet.PositionalEmbedding(dim: int, max_length: int = 10000)

Bases: Module

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

embedding: Tensor

forward(x: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

static make_embedding(dim: int, max_length: int = 10000) → Tensor

class ml.models.architectures.unet.FFN(in_dim: int, embed_dim: int)

Bases: Module

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

forward(x: Tensor, t: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class ml.models.architectures.unet.BasicBlock(in_c: int, out_c: int, embed_c: int | None = None, act: Literal['no_act', 'relu', 'relu6', 'relu2', 'clamp6', 'leaky_relu', 'elu', 'celu', 'selu', 'gelu', 'gelu_fast', 'sigmoid', 'log_sigmoid', 'hard_sigomid', 'tanh', 'softsign', 'softplus', 'silu', 'mish', 'swish', 'hard_swish', 'soft_shrink', 'hard_shrink', 'tanh_shrink', 'soft_sign', 'relu_squared', 'laplace'] = 'relu', norm: Literal['no_norm', 'batch', 'batch_affine', 'instance', 'instance_affine', 'group', 'group_affine', 'layer', 'layer_affine'] = 'batch_affine')

Bases: Module

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

forward(x: Tensor, embedding: Tensor | None = None) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class ml.models.architectures.unet.SelfAttention2d(dim: int, num_heads: int = 8, dropout_prob: float = 0.1)

Bases: Module

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

forward(x: Tensor) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class ml.models.architectures.unet.UNet(in_dim: int, embed_dim: int, dim_scales: Sequence[int], input_embedding_dim: int | None = None)

Bases: Module

Defines a general-purpose UNet model.

Parameters:

• in_dim – Number of input dimensions.

• embed_dim – Embedding dimension.

• dim_scales – List of dimension scales.

• input_embedding_dim – The input embedding dimension, if an input embedding is used (for example, when conditioning on time, or some class embedding).

Inputs:

x: Input tensor of shape (batch_size, in_dim, height, width). t: Time tensor of shape (batch_size) if use_time is True

and None otherwise.

c: Class tensor of shape (batch_size, class_dim) if use_class

is True and None otherwise.

Outputs:

x: Output tensor of shape (batch_size, in_dim, height, width).

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

forward(x: Tensor, embedding: Tensor | None = None) → Tensor

Defines the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.