Luis Oala
first test
d62813f
"""
Various utilities for neural networks.
"""
import math
import torch as th
import torch.nn as nn
import torch.nn.functional as F
class GroupNorm32(nn.GroupNorm):
def __init__(self, num_groups, num_channels, swish, eps=1e-5):
super().__init__(num_groups=num_groups, num_channels=num_channels, eps=eps)
self.swish = swish
def forward(self, x):
y = super().forward(x.float()).to(x.dtype)
if self.swish == 1.0:
y = F.silu(y)
elif self.swish:
y = y * F.sigmoid(y * float(self.swish))
return y
def conv_nd(dims, *args, **kwargs):
"""
Create a 1D, 2D, or 3D convolution module.
"""
if dims == 1:
return nn.Conv1d(*args, **kwargs)
elif dims == 2:
return nn.Conv2d(*args, **kwargs)
elif dims == 3:
return nn.Conv3d(*args, **kwargs)
raise ValueError(f"unsupported dimensions: {dims}")
def linear(*args, **kwargs):
"""
Create a linear module.
"""
return nn.Linear(*args, **kwargs)
def avg_pool_nd(dims, *args, **kwargs):
"""
Create a 1D, 2D, or 3D average pooling module.
"""
if dims == 1:
return nn.AvgPool1d(*args, **kwargs)
elif dims == 2:
return nn.AvgPool2d(*args, **kwargs)
elif dims == 3:
return nn.AvgPool3d(*args, **kwargs)
raise ValueError(f"unsupported dimensions: {dims}")
def zero_module(module):
"""
Zero out the parameters of a module and return it.
"""
for p in module.parameters():
p.detach().zero_()
return module
def scale_module(module, scale):
"""
Scale the parameters of a module and return it.
"""
for p in module.parameters():
p.detach().mul_(scale)
return module
def normalization(channels, swish=0.0):
"""
Make a standard normalization layer, with an optional swish activation.
:param channels: number of input channels.
:return: an nn.Module for normalization.
"""
return GroupNorm32(num_channels=channels, num_groups=32, swish=swish)
def timestep_embedding(timesteps, dim, max_period=10000):
"""
Create sinusoidal timestep embeddings.
:param timesteps: a 1-D Tensor of N indices, one per batch element.
These may be fractional.
:param dim: the dimension of the output.
:param max_period: controls the minimum frequency of the embeddings.
:return: an [N x dim] Tensor of positional embeddings.
"""
half = dim // 2
freqs = th.exp(
-math.log(max_period) * th.arange(start=0, end=half, dtype=th.float32) / half
).to(device=timesteps.device)
args = timesteps[:, None].float() * freqs[None]
embedding = th.cat([th.cos(args), th.sin(args)], dim=-1)
if dim % 2:
embedding = th.cat([embedding, th.zeros_like(embedding[:, :1])], dim=-1)
return embedding