1044 lines
34 KiB
Python
1044 lines
34 KiB
Python
import os
|
|
training_mode = os.getenv("training_mode")
|
|
|
|
# pytorch_diffusion + derived encoder decoder
|
|
import logging
|
|
import math
|
|
from typing import Any, Callable, Optional
|
|
|
|
import numpy as np
|
|
import torch
|
|
if training_mode == "npu":
|
|
import torch_npu
|
|
from torch_npu.contrib import transfer_to_npu
|
|
import torch.nn as nn
|
|
from inspect import isfunction
|
|
from einops import rearrange, repeat
|
|
from packaging import version
|
|
from typing import Tuple
|
|
from torch import Tensor
|
|
from typing import Union, Tuple
|
|
|
|
logpy = logging.getLogger(__name__)
|
|
|
|
try:
|
|
import xformers
|
|
import xformers.ops
|
|
|
|
XFORMERS_IS_AVAILABLE = True
|
|
except:
|
|
XFORMERS_IS_AVAILABLE = False
|
|
logpy.warning("no module 'xformers'. Processing without...")
|
|
|
|
|
|
def exists(val):
|
|
return val is not None
|
|
|
|
def default(val, d):
|
|
if exists(val):
|
|
return val
|
|
return d() if isfunction(d) else d
|
|
|
|
def get_timestep_embedding(timesteps, embedding_dim):
|
|
"""
|
|
This matches the implementation in Denoising Diffusion Probabilistic Models:
|
|
From Fairseq.
|
|
Build sinusoidal embeddings.
|
|
This matches the implementation in tensor2tensor, but differs slightly
|
|
from the description in Section 3.5 of "Attention Is All You Need".
|
|
"""
|
|
assert len(timesteps.shape) == 1
|
|
|
|
half_dim = embedding_dim // 2
|
|
emb = math.log(10000) / (half_dim - 1)
|
|
emb = torch.exp(torch.arange(half_dim, dtype=torch.float32) * -emb)
|
|
emb = emb.to(device=timesteps.device)
|
|
emb = timesteps.float()[:, None] * emb[None, :]
|
|
emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
|
|
if embedding_dim % 2 == 1: # zero pad
|
|
emb = torch.nn.functional.pad(emb, (0, 1, 0, 0))
|
|
return emb
|
|
|
|
|
|
class LinearAttention(nn.Module):
|
|
def __init__(self, dim, heads=4, dim_head=32):
|
|
super().__init__()
|
|
self.heads = heads
|
|
hidden_dim = dim_head * heads
|
|
self.to_qkv = nn.Conv2d(dim, hidden_dim * 3, 1, bias=False)
|
|
self.to_out = nn.Conv2d(hidden_dim, dim, 1)
|
|
|
|
def forward(self, x):
|
|
b, c, h, w = x.shape
|
|
qkv = self.to_qkv(x)
|
|
q, k, v = rearrange(
|
|
qkv, "b (qkv heads c) h w -> qkv b heads c (h w)", heads=self.heads, qkv=3
|
|
)
|
|
k = k.softmax(dim=-1)
|
|
context = torch.einsum("bhdn,bhen->bhde", k, v)
|
|
out = torch.einsum("bhde,bhdn->bhen", context, q)
|
|
out = rearrange(
|
|
out, "b heads c (h w) -> b (heads c) h w", heads=self.heads, h=h, w=w
|
|
)
|
|
return self.to_out(out)
|
|
|
|
class MemoryEfficientCrossAttention(nn.Module):
|
|
# https://github.com/MatthieuTPHR/diffusers011/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers011/models/attention.py#L223
|
|
def __init__(
|
|
self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.0, **kwargs
|
|
):
|
|
super().__init__()
|
|
logpy.debug(
|
|
f"Setting up {self.__class__.__name__}. Query dim is {query_dim}, "
|
|
f"context_dim is {context_dim} and using {heads} heads with a "
|
|
f"dimension of {dim_head}."
|
|
)
|
|
inner_dim = dim_head * heads
|
|
context_dim = default(context_dim, query_dim)
|
|
|
|
self.heads = heads
|
|
self.dim_head = dim_head
|
|
|
|
self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
|
|
self.to_k = nn.Linear(context_dim, inner_dim, bias=False)
|
|
self.to_v = nn.Linear(context_dim, inner_dim, bias=False)
|
|
|
|
self.to_out = nn.Sequential(
|
|
nn.Linear(inner_dim, query_dim), nn.Dropout(dropout)
|
|
)
|
|
self.attention_op: Optional[Any] = None
|
|
|
|
def forward(
|
|
self,
|
|
x,
|
|
context=None,
|
|
mask=None,
|
|
additional_tokens=None,
|
|
n_times_crossframe_attn_in_self=0,
|
|
):
|
|
if additional_tokens is not None:
|
|
# get the number of masked tokens at the beginning of the output sequence
|
|
n_tokens_to_mask = additional_tokens.shape[1]
|
|
# add additional token
|
|
x = torch.cat([additional_tokens, x], dim=1)
|
|
q = self.to_q(x)
|
|
context = default(context, x)
|
|
k = self.to_k(context)
|
|
v = self.to_v(context)
|
|
|
|
if n_times_crossframe_attn_in_self:
|
|
# reprogramming cross-frame attention as in https://arxiv.org/abs/2303.13439
|
|
assert x.shape[0] % n_times_crossframe_attn_in_self == 0
|
|
# n_cp = x.shape[0]//n_times_crossframe_attn_in_self
|
|
k = repeat(
|
|
k[::n_times_crossframe_attn_in_self],
|
|
"b ... -> (b n) ...",
|
|
n=n_times_crossframe_attn_in_self,
|
|
)
|
|
v = repeat(
|
|
v[::n_times_crossframe_attn_in_self],
|
|
"b ... -> (b n) ...",
|
|
n=n_times_crossframe_attn_in_self,
|
|
)
|
|
|
|
b, _, _ = q.shape
|
|
q, k, v = map(
|
|
lambda t: t.unsqueeze(3)
|
|
.reshape(b, t.shape[1], self.heads, self.dim_head)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(b * self.heads, t.shape[1], self.dim_head)
|
|
.contiguous(),
|
|
(q, k, v),
|
|
)
|
|
|
|
# actually compute the attention, what we cannot get enough of
|
|
if version.parse(xformers.__version__) >= version.parse("0.0.21"):
|
|
# NOTE: workaround for
|
|
# https://github.com/facebookresearch/xformers/issues/845
|
|
max_bs = 32768
|
|
N = q.shape[0]
|
|
n_batches = math.ceil(N / max_bs)
|
|
out = list()
|
|
for i_batch in range(n_batches):
|
|
batch = slice(i_batch * max_bs, (i_batch + 1) * max_bs)
|
|
out.append(
|
|
xformers.ops.memory_efficient_attention(
|
|
q[batch],
|
|
k[batch],
|
|
v[batch],
|
|
attn_bias=None,
|
|
op=self.attention_op,
|
|
)
|
|
)
|
|
out = torch.cat(out, 0)
|
|
else:
|
|
out = xformers.ops.memory_efficient_attention(
|
|
q, k, v, attn_bias=None, op=self.attention_op
|
|
)
|
|
|
|
# TODO: Use this directly in the attention operation, as a bias
|
|
if exists(mask):
|
|
raise NotImplementedError
|
|
out = (
|
|
out.unsqueeze(0)
|
|
.reshape(b, self.heads, out.shape[1], self.dim_head)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(b, out.shape[1], self.heads * self.dim_head)
|
|
)
|
|
if additional_tokens is not None:
|
|
# remove additional token
|
|
out = out[:, n_tokens_to_mask:]
|
|
return self.to_out(out)
|
|
|
|
def nonlinearity(x):
|
|
# swish
|
|
return x * torch.sigmoid(x)
|
|
|
|
|
|
def Normalize(in_channels, num_groups=32):
|
|
return torch.nn.GroupNorm(
|
|
num_groups=num_groups, num_channels=in_channels, eps=1e-5, affine=True
|
|
)
|
|
|
|
|
|
class Upsample3D(nn.Module):
|
|
def __init__(self, in_channels, with_conv, up_time=False, causal=False):
|
|
super().__init__()
|
|
self.with_conv = with_conv
|
|
assert with_conv is True
|
|
self.up_time = 2 if up_time else 1
|
|
self.causal = causal
|
|
if self.with_conv:
|
|
self.conv = torch.nn.Conv3d(
|
|
in_channels,
|
|
in_channels * self.up_time,
|
|
kernel_size=3,
|
|
stride=1,
|
|
padding=0,
|
|
)
|
|
|
|
def forward(self, x):
|
|
ori_dtype = x.dtype
|
|
if x.dtype == torch.bfloat16:
|
|
x = x.to(torch.float16)
|
|
x = torch.nn.functional.interpolate(
|
|
x, scale_factor=(1.0, 2.0, 2.0), mode="nearest"
|
|
)
|
|
|
|
if self.with_conv:
|
|
x = torch.nn.functional.pad(x, (1, 1, 1, 1, 0, 0), mode="constant", value=0)
|
|
if not self.causal:
|
|
x = torch.nn.functional.pad(x, (0, 0, 0, 0, 1, 1), mode="replicate")
|
|
else:
|
|
x = torch.nn.functional.pad(x, (0, 0, 0, 0, 2, 0), mode="replicate")
|
|
x = x.to(ori_dtype)
|
|
x = self.conv(x)
|
|
x = rearrange(x, "b (n c) t h w -> b c (t n) h w", n=self.up_time)
|
|
if self.up_time != 1:
|
|
x = x[:, :, 1:, :, :]
|
|
else:
|
|
x = x.to(ori_dtype)
|
|
return x
|
|
|
|
|
|
class Downsample3D(nn.Module):
|
|
def __init__(self, in_channels, with_conv, down_time=False):
|
|
super().__init__()
|
|
self.with_conv = with_conv
|
|
assert with_conv is True
|
|
self.down_time = down_time
|
|
self.stride = 2 if self.down_time else (1, 2, 2)
|
|
if self.with_conv:
|
|
# no asymmetric padding in torch conv, must do it ourselves
|
|
self.conv = torch.nn.Conv3d(
|
|
in_channels, in_channels, kernel_size=3, stride=self.stride, padding=0
|
|
)
|
|
|
|
def forward(self, x):
|
|
if self.with_conv:
|
|
x = torch.nn.functional.pad(x, (0, 1, 0, 1, 0, 0), mode="constant", value=0)
|
|
if x.dtype == torch.bfloat16:
|
|
x = torch.nn.functional.pad(
|
|
x.to(torch.float16), (0, 0, 0, 0, 2, 0), mode="replicate"
|
|
).to(torch.bfloat16)
|
|
else:
|
|
x = torch.nn.functional.pad(x, (0, 0, 0, 0, 2, 0), mode="replicate")
|
|
x = self.conv(x)
|
|
else:
|
|
x = torch.nn.functional.avg_pool3d(x, kernel_size=self.stride, stride=2)
|
|
return x
|
|
|
|
|
|
class CausalConv3d(torch.nn.Conv3d):
|
|
def __init__(
|
|
self,
|
|
in_channels: int,
|
|
out_channels: int,
|
|
kernel_size: Union[int, Tuple[int, int, int]],
|
|
stride: Union[int, Tuple[int, int, int]] = 1,
|
|
padding: Union[str, int, Tuple[int, int, int]] = 0,
|
|
dilation: Union[int, Tuple[int, int, int]] = 1,
|
|
groups: int = 1,
|
|
bias: bool = True,
|
|
padding_mode: str = "replicate",
|
|
device=None,
|
|
dtype=None,
|
|
) -> None:
|
|
super().__init__(
|
|
in_channels,
|
|
out_channels,
|
|
kernel_size,
|
|
stride,
|
|
0,
|
|
dilation,
|
|
groups,
|
|
bias,
|
|
"zeros",
|
|
device,
|
|
dtype,
|
|
)
|
|
assert isinstance(padding, int)
|
|
self.custom_padding = padding
|
|
self.time_padding_mode = padding_mode
|
|
|
|
def forward(self, input: Tensor) -> Tensor:
|
|
ori_dtype = input.dtype
|
|
input = input.to(dtype=torch.float32)
|
|
input = torch.nn.functional.pad(
|
|
input,
|
|
(
|
|
self.custom_padding,
|
|
self.custom_padding,
|
|
self.custom_padding,
|
|
self.custom_padding,
|
|
0,
|
|
0,
|
|
),
|
|
mode="constant",
|
|
value=0,
|
|
)
|
|
|
|
if input.dtype == torch.bfloat16:
|
|
input = torch.nn.functional.pad(
|
|
input.to(torch.float16),
|
|
(0, 0, 0, 0, 2 * self.custom_padding, 0),
|
|
mode=self.time_padding_mode,
|
|
).to(torch.bfloat16)
|
|
else:
|
|
input = torch.nn.functional.pad(
|
|
input,
|
|
(0, 0, 0, 0, 2 * self.custom_padding, 0),
|
|
mode=self.time_padding_mode,
|
|
)
|
|
input = input.to(ori_dtype)
|
|
return super().forward(input)
|
|
|
|
|
|
class Conv2dWithExtraDim(torch.nn.Conv2d):
|
|
def forward(self, input: Tensor) -> Tensor:
|
|
if input.dim() == 5:
|
|
b, c, t, h, w = input.shape
|
|
input = rearrange(input, "b c t h w -> (b t) c h w")
|
|
output = super().forward(input)
|
|
output = rearrange(output, "(b t) c h w -> b c t h w", b=b, t=t)
|
|
else:
|
|
output = super().forward(input)
|
|
return output
|
|
|
|
|
|
class ResnetBlock3D(nn.Module):
|
|
def __init__(
|
|
self,
|
|
*,
|
|
in_channels,
|
|
out_channels=None,
|
|
conv_shortcut=False,
|
|
dropout,
|
|
temb_channels=512,
|
|
use_3d_conv=True,
|
|
half_3d=True,
|
|
causal=False,
|
|
):
|
|
super().__init__()
|
|
self.in_channels = in_channels
|
|
out_channels = in_channels if out_channels is None else out_channels
|
|
self.out_channels = out_channels
|
|
self.use_conv_shortcut = conv_shortcut
|
|
conv_cls = CausalConv3d if causal else nn.Conv3d
|
|
conv_cls = conv_cls if use_3d_conv else Conv2dWithExtraDim
|
|
|
|
self.norm1 = Normalize(in_channels)
|
|
self.conv1 = conv_cls(
|
|
in_channels, out_channels, kernel_size=3, stride=1, padding=1
|
|
)
|
|
if temb_channels > 0:
|
|
self.temb_proj = torch.nn.Linear(temb_channels, out_channels)
|
|
self.norm2 = Normalize(out_channels)
|
|
self.dropout = torch.nn.Dropout(dropout)
|
|
if half_3d:
|
|
self.conv2 = Conv2dWithExtraDim(
|
|
out_channels, out_channels, kernel_size=3, stride=1, padding=1
|
|
)
|
|
else:
|
|
self.conv2 = conv_cls(
|
|
out_channels, out_channels, kernel_size=3, stride=1, padding=1
|
|
)
|
|
if self.in_channels != self.out_channels:
|
|
if self.use_conv_shortcut:
|
|
self.conv_shortcut = conv_cls(
|
|
in_channels, out_channels, kernel_size=3, stride=1, padding=1
|
|
)
|
|
else:
|
|
self.nin_shortcut = conv_cls(
|
|
in_channels, out_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
|
|
def forward(self, x, temb=None):
|
|
h = x
|
|
h = self.norm1(h)
|
|
h = nonlinearity(h)
|
|
h = self.conv1(h)
|
|
|
|
if temb is not None:
|
|
h = h + self.temb_proj(nonlinearity(temb))[:, :, None, None]
|
|
|
|
h = self.norm2(h)
|
|
h = nonlinearity(h)
|
|
h = self.dropout(h)
|
|
h = self.conv2(h)
|
|
|
|
if self.in_channels != self.out_channels:
|
|
if self.use_conv_shortcut:
|
|
x = self.conv_shortcut(x)
|
|
else:
|
|
x = self.nin_shortcut(x)
|
|
|
|
return x + h
|
|
|
|
|
|
class LinAttnBlock(LinearAttention):
|
|
"""to match AttnBlock usage"""
|
|
|
|
def __init__(self, in_channels):
|
|
super().__init__(dim=in_channels, heads=1, dim_head=in_channels)
|
|
|
|
def forward(self, x):
|
|
b, c, t, h, w = x.shape
|
|
x = rearrange(x, "b c t h w -> (b t) c h w")
|
|
x = super().forward(x)
|
|
x = rearrange(x, "(b t) c h w -> b c t h w", b=b, t=t)
|
|
return x
|
|
|
|
|
|
class AttnBlock(nn.Module):
|
|
def __init__(self, in_channels):
|
|
super().__init__()
|
|
self.in_channels = in_channels
|
|
|
|
self.norm = Normalize(in_channels)
|
|
self.q = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.k = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.v = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.proj_out = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
|
|
def attention(self, h_: torch.Tensor) -> torch.Tensor:
|
|
h_ = self.norm(h_)
|
|
q = self.q(h_)
|
|
k = self.k(h_)
|
|
v = self.v(h_)
|
|
|
|
b, c, h, w = q.shape
|
|
q, k, v = map(
|
|
lambda x: rearrange(x, "b c h w -> b 1 (h w) c").contiguous(), (q, k, v)
|
|
)
|
|
h_ = torch.nn.functional.scaled_dot_product_attention(
|
|
q, k, v
|
|
) # scale is dim ** -0.5 per default
|
|
# compute attention
|
|
|
|
return rearrange(h_, "b 1 (h w) c -> b c h w", h=h, w=w, c=c, b=b)
|
|
|
|
def forward(self, x, **kwargs):
|
|
h_ = x
|
|
b, c, t, h, w = h_.shape
|
|
h_ = rearrange(h_, "b c t h w -> (b t) c h w")
|
|
h_ = self.attention(h_)
|
|
h_ = self.proj_out(h_)
|
|
h_ = rearrange(h_, "(b t) c h w -> b c t h w", b=b, t=t)
|
|
return x + h_
|
|
|
|
|
|
class MemoryEfficientAttnBlock(nn.Module):
|
|
"""
|
|
Uses xformers efficient implementation,
|
|
see https://github.com/MatthieuTPHR/diffusers011/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers011/models/attention.py#L223
|
|
Note: this is a single-head self-attention operation
|
|
"""
|
|
|
|
#
|
|
def __init__(self, in_channels):
|
|
super().__init__()
|
|
self.in_channels = in_channels
|
|
|
|
self.norm = Normalize(in_channels)
|
|
self.q = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.k = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.v = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.proj_out = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.attention_op: Optional[Any] = None
|
|
|
|
def attention(self, h_: torch.Tensor) -> torch.Tensor:
|
|
h_ = self.norm(h_)
|
|
q = self.q(h_)
|
|
k = self.k(h_)
|
|
v = self.v(h_)
|
|
|
|
# compute attention
|
|
B, C, H, W = q.shape
|
|
q, k, v = map(lambda x: rearrange(x, "b c h w -> b (h w) c"), (q, k, v))
|
|
|
|
q, k, v = map(
|
|
lambda t: t.unsqueeze(3)
|
|
.reshape(B, t.shape[1], 1, C)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(B * 1, t.shape[1], C)
|
|
.contiguous(),
|
|
(q, k, v),
|
|
)
|
|
if training_mode == "npu":
|
|
heads = 8
|
|
out = torch_npu.npu_fusion_attention(
|
|
q, k, v, heads, input_layout="BSH",
|
|
pse=None,
|
|
atten_mask=None,
|
|
pre_tockens=2147483647,
|
|
next_tockens=2147483647,
|
|
keep_prob=1.,
|
|
sync=False,
|
|
inner_precise=0,
|
|
)[0]
|
|
else:
|
|
out = xformers.ops.memory_efficient_attention(
|
|
q, k, v, attn_bias=None, op=self.attention_op
|
|
)
|
|
out = (
|
|
out.unsqueeze(0)
|
|
.reshape(B, 1, out.shape[1], C)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(B, out.shape[1], C)
|
|
)
|
|
return rearrange(out, "b (h w) c -> b c h w", b=B, h=H, w=W, c=C)
|
|
|
|
def forward(self, x, **kwargs):
|
|
h_ = x
|
|
b, c, t, h, w = h_.shape
|
|
h_ = rearrange(h_, "b c t h w -> (b t) c h w")
|
|
h_ = self.attention(h_)
|
|
h_ = self.proj_out(h_)
|
|
h_ = rearrange(h_, "(b t) c h w -> b c t h w", b=b, t=t)
|
|
return x + h_
|
|
|
|
|
|
class MemoryEfficientAttnVideoBlock(nn.Module):
|
|
"""
|
|
Uses xformers efficient implementation,
|
|
see https://github.com/MatthieuTPHR/diffusers011/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers011/models/attention.py#L223
|
|
Note: this is a single-head self-attention operation
|
|
"""
|
|
|
|
#
|
|
def __init__(self, in_channels):
|
|
super().__init__()
|
|
self.in_channels = in_channels
|
|
|
|
self.norm = Normalize(in_channels)
|
|
self.q = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.k = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.v = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.proj_out = torch.nn.Conv2d(
|
|
in_channels, in_channels, kernel_size=1, stride=1, padding=0
|
|
)
|
|
self.attention_op: Optional[Any] = None
|
|
|
|
self.q_t = torch.nn.Linear(in_channels, in_channels)
|
|
self.k_t = torch.nn.Linear(in_channels, in_channels)
|
|
self.v_t = torch.nn.Linear(in_channels, in_channels)
|
|
self.proj_out_t = torch.nn.Linear(in_channels, in_channels)
|
|
self.norm_t = torch.nn.LayerNorm(in_channels)
|
|
|
|
def attention_t(self, h_: torch.Tensor) -> torch.Tensor:
|
|
# (b h w) t c
|
|
h_ = self.norm_t(h_)
|
|
# compute attention
|
|
q = self.q_t(h_).contiguous()
|
|
k = self.k_t(h_).contiguous()
|
|
v = self.v_t(h_).contiguous()
|
|
if training_mode == "npu":
|
|
heads = 8
|
|
out = torch_npu.npu_fusion_attention(
|
|
q, k, v, heads, input_layout="BSH",
|
|
pse=None,
|
|
atten_mask=None,
|
|
pre_tockens=2147483647,
|
|
next_tockens=2147483647,
|
|
keep_prob=1.,
|
|
sync=False,
|
|
inner_precise=0,
|
|
)[0]
|
|
else:
|
|
out = xformers.ops.memory_efficient_attention(
|
|
q, k, v, attn_bias=None, op=self.attention_op
|
|
)
|
|
out = self.proj_out_t(out)
|
|
|
|
return out
|
|
|
|
def attention(self, h_: torch.Tensor) -> torch.Tensor:
|
|
h_ = self.norm(h_)
|
|
q = self.q(h_)
|
|
k = self.k(h_)
|
|
v = self.v(h_)
|
|
|
|
# compute attention
|
|
B, C, H, W = q.shape
|
|
q, k, v = map(lambda x: rearrange(x, "b c h w -> b (h w) c"), (q, k, v))
|
|
|
|
q, k, v = map(
|
|
lambda t: t.unsqueeze(3)
|
|
.reshape(B, t.shape[1], 1, C)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(B * 1, t.shape[1], C)
|
|
.contiguous(),
|
|
(q, k, v),
|
|
)
|
|
if training_mode == "npu":
|
|
heads = 8
|
|
out = torch_npu.npu_fusion_attention(
|
|
q, k, v, heads, input_layout="BSH",
|
|
pse=None,
|
|
atten_mask=None,
|
|
pre_tockens=2147483647,
|
|
next_tockens=2147483647,
|
|
keep_prob=1.,
|
|
sync=False,
|
|
inner_precise=0,
|
|
)[0]
|
|
else:
|
|
out = xformers.ops.memory_efficient_attention(
|
|
q, k, v, attn_bias=None, op=self.attention_op
|
|
)
|
|
out = (
|
|
out.unsqueeze(0)
|
|
.reshape(B, 1, out.shape[1], C)
|
|
.permute(0, 2, 1, 3)
|
|
.reshape(B, out.shape[1], C)
|
|
)
|
|
return rearrange(out, "b (h w) c -> b c h w", b=B, h=H, w=W, c=C)
|
|
|
|
def forward(self, x, **kwargs):
|
|
h_ = x
|
|
b, c, t, h, w = h_.shape
|
|
h_ = rearrange(h_, "b c t h w -> (b t) c h w")
|
|
h_ = self.attention(h_)
|
|
h_ = self.proj_out(h_)
|
|
# h_ = rearrange(h_, "(b t) c h w -> b c t h w", b=b, t=t)
|
|
h_ = rearrange(h_, "(b t) c h w -> (b h w) t c", b=b, t=t)
|
|
h_ = self.attention_t(h_)
|
|
h_ = rearrange(h_, "(b h w) t c -> b c t h w", b=b, h=h, w=w)
|
|
return x + h_
|
|
|
|
|
|
class MemoryEfficientCrossAttentionWrapper(MemoryEfficientCrossAttention):
|
|
def forward(self, x, context=None, mask=None, **unused_kwargs):
|
|
b, c, h, w = x.shape
|
|
x = rearrange(x, "b c h w -> b (h w) c")
|
|
out = super().forward(x, context=context, mask=mask)
|
|
out = rearrange(out, "b (h w) c -> b c h w", h=h, w=w, c=c)
|
|
return x + out
|
|
|
|
|
|
def make_attn(in_channels, attn_type="vanilla", attn_kwargs=None):
|
|
assert attn_type in [
|
|
"vanilla",
|
|
"vanilla-xformers",
|
|
"spatial-temporal-xformer",
|
|
"memory-efficient-cross-attn",
|
|
"linear",
|
|
"none",
|
|
], f"attn_type {attn_type} unknown"
|
|
if (
|
|
version.parse(torch.__version__) < version.parse("2.0.0")
|
|
and attn_type != "none"
|
|
):
|
|
assert XFORMERS_IS_AVAILABLE, (
|
|
f"We do not support vanilla attention in {torch.__version__} anymore, "
|
|
f"as it is too expensive. Please install xformers via e.g. 'pip install xformers==0.0.16'"
|
|
)
|
|
attn_type = "vanilla-xformers"
|
|
logpy.info(f"making attention of type '{attn_type}' with {in_channels} in_channels")
|
|
if attn_type == "vanilla":
|
|
assert attn_kwargs is None
|
|
return AttnBlock(in_channels)
|
|
elif attn_type == "vanilla-xformers":
|
|
logpy.info(
|
|
f"building MemoryEfficientAttnBlock with {in_channels} in_channels..."
|
|
)
|
|
return MemoryEfficientAttnBlock(in_channels)
|
|
elif type == "memory-efficient-cross-attn":
|
|
attn_kwargs["query_dim"] = in_channels
|
|
return MemoryEfficientCrossAttentionWrapper(**attn_kwargs)
|
|
elif attn_type == "none":
|
|
return nn.Identity(in_channels)
|
|
elif attn_type == "spatial-temporal-xformer":
|
|
return MemoryEfficientAttnVideoBlock(in_channels)
|
|
else:
|
|
return LinAttnBlock(in_channels)
|
|
|
|
|
|
class Encoder(nn.Module):
|
|
def __init__(
|
|
self,
|
|
*,
|
|
ch,
|
|
out_ch,
|
|
ch_mult=(1, 2, 4, 8),
|
|
num_res_blocks,
|
|
attn_resolutions,
|
|
dropout=0.0,
|
|
resamp_with_conv=True,
|
|
in_channels,
|
|
z_channels,
|
|
resolution=256,
|
|
double_z=True,
|
|
use_linear_attn=False,
|
|
attn_type="vanilla",
|
|
use_3d_conv=True,
|
|
half_3d=True,
|
|
causal=True,
|
|
half_t_mult=True,
|
|
gradient_checkpointing=True,
|
|
**ignore_kwargs,
|
|
):
|
|
super().__init__()
|
|
if use_linear_attn:
|
|
attn_type = "linear"
|
|
self.ch = ch
|
|
self.temb_ch = 0
|
|
self.num_resolutions = len(ch_mult)
|
|
self.num_res_blocks = num_res_blocks
|
|
self.resolution = resolution
|
|
self.in_channels = in_channels
|
|
self.gradient_checkpointing = gradient_checkpointing
|
|
|
|
conv_cls = CausalConv3d if causal else nn.Conv3d
|
|
conv_cls = conv_cls if use_3d_conv else Conv2dWithExtraDim
|
|
|
|
# downsampling
|
|
self.conv_in = conv_cls(
|
|
in_channels, self.ch, kernel_size=3, stride=1, padding=1
|
|
)
|
|
|
|
curr_res = resolution
|
|
in_ch_mult = (1,) + tuple(ch_mult)
|
|
self.in_ch_mult = in_ch_mult
|
|
self.down = nn.ModuleList()
|
|
for i_level in range(self.num_resolutions):
|
|
block = nn.ModuleList()
|
|
attn = nn.ModuleList()
|
|
block_in = ch * in_ch_mult[i_level]
|
|
block_out = ch * ch_mult[i_level]
|
|
for i_block in range(self.num_res_blocks):
|
|
block.append(
|
|
ResnetBlock3D(
|
|
in_channels=block_in,
|
|
out_channels=block_out,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
)
|
|
block_in = block_out
|
|
if curr_res in attn_resolutions:
|
|
attn.append(make_attn(block_in, attn_type=attn_type))
|
|
down = nn.Module()
|
|
down.block = block
|
|
down.attn = attn
|
|
if i_level != self.num_resolutions - 1:
|
|
if half_t_mult:
|
|
down_time = i_level % 2 == 0
|
|
else:
|
|
down_time = True
|
|
down.downsample = Downsample3D(block_in, resamp_with_conv, down_time)
|
|
curr_res = curr_res // 2
|
|
self.down.append(down)
|
|
|
|
# middle
|
|
self.mid = nn.Module()
|
|
self.mid.block_1 = ResnetBlock3D(
|
|
in_channels=block_in,
|
|
out_channels=block_in,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
self.mid.attn_1 = make_attn(block_in, attn_type=attn_type)
|
|
self.mid.block_2 = ResnetBlock3D(
|
|
in_channels=block_in,
|
|
out_channels=block_in,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
|
|
# end
|
|
self.norm_out = Normalize(block_in)
|
|
self.conv_out = conv_cls(
|
|
block_in,
|
|
2 * z_channels if double_z else z_channels,
|
|
kernel_size=3,
|
|
stride=1,
|
|
padding=1,
|
|
)
|
|
|
|
def forward(self, x):
|
|
# timestep embedding
|
|
temb = None
|
|
|
|
# downsampling
|
|
hs = [self.conv_in(x)]
|
|
for i_level in range(self.num_resolutions):
|
|
for i_block in range(self.num_res_blocks):
|
|
if self.gradient_checkpointing and self.training:
|
|
h = torch.utils.checkpoint.checkpoint(
|
|
self.down[i_level].block[i_block].__call__,
|
|
hs[-1],
|
|
temb,
|
|
use_reentrant=False,
|
|
)
|
|
else:
|
|
h = self.down[i_level].block[i_block](hs[-1], temb)
|
|
if len(self.down[i_level].attn) > 0:
|
|
h = self.down[i_level].attn[i_block](h)
|
|
hs.append(h)
|
|
if i_level != self.num_resolutions - 1:
|
|
hs.append(self.down[i_level].downsample(hs[-1]))
|
|
|
|
# middle
|
|
h = hs[-1]
|
|
h = self.mid.block_1(h, temb)
|
|
h = self.mid.attn_1(h)
|
|
h = self.mid.block_2(h, temb)
|
|
|
|
# end
|
|
h = self.norm_out(h)
|
|
h = nonlinearity(h)
|
|
h = self.conv_out(h)
|
|
return h
|
|
|
|
|
|
class Decoder(nn.Module):
|
|
def __init__(
|
|
self,
|
|
*,
|
|
ch,
|
|
out_ch,
|
|
ch_mult=(1, 2, 4, 8),
|
|
num_res_blocks,
|
|
attn_resolutions,
|
|
dropout=0.0,
|
|
resamp_with_conv=True,
|
|
in_channels,
|
|
z_channels,
|
|
resolution=256,
|
|
give_pre_end=False,
|
|
tanh_out=False,
|
|
use_linear_attn=False,
|
|
attn_type="vanilla",
|
|
use_3d_conv=True,
|
|
half_3d=True,
|
|
causal=True,
|
|
half_t_mult=True,
|
|
gradient_checkpointing=True,
|
|
**ignorekwargs,
|
|
):
|
|
super().__init__()
|
|
if use_linear_attn:
|
|
attn_type = "linear"
|
|
self.ch = ch
|
|
self.temb_ch = 0
|
|
self.num_resolutions = len(ch_mult)
|
|
self.num_res_blocks = num_res_blocks
|
|
self.resolution = resolution
|
|
self.in_channels = in_channels
|
|
self.give_pre_end = give_pre_end
|
|
self.tanh_out = tanh_out
|
|
self.gradient_checkpointing = gradient_checkpointing
|
|
|
|
# compute in_ch_mult, block_in and curr_res at lowest res
|
|
in_ch_mult = (1,) + tuple(ch_mult)
|
|
block_in = ch * ch_mult[self.num_resolutions - 1]
|
|
curr_res = resolution // 2 ** (self.num_resolutions - 1)
|
|
self.z_shape = (1, z_channels, curr_res, curr_res)
|
|
logpy.info(
|
|
"Working with z of shape {} = {} dimensions.".format(
|
|
self.z_shape, np.prod(self.z_shape)
|
|
)
|
|
)
|
|
self.causal = causal
|
|
self.use_3d_conv = use_3d_conv
|
|
|
|
make_attn_cls = self._make_attn()
|
|
make_resblock_cls = self._make_resblock()
|
|
make_conv_cls = self._make_conv()
|
|
# z to block_in
|
|
self.conv_in = make_conv_cls(
|
|
z_channels, block_in, kernel_size=3, stride=1, padding=1
|
|
)
|
|
|
|
# middle
|
|
self.mid = nn.Module()
|
|
self.mid.block_1 = make_resblock_cls(
|
|
in_channels=block_in,
|
|
out_channels=block_in,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
self.mid.attn_1 = make_attn_cls(block_in, attn_type=attn_type)
|
|
self.mid.block_2 = make_resblock_cls(
|
|
in_channels=block_in,
|
|
out_channels=block_in,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
|
|
# upsampling
|
|
self.up = nn.ModuleList()
|
|
for i_level in reversed(range(self.num_resolutions)):
|
|
block = nn.ModuleList()
|
|
attn = nn.ModuleList()
|
|
block_out = ch * ch_mult[i_level]
|
|
for i_block in range(self.num_res_blocks + 1):
|
|
block.append(
|
|
make_resblock_cls(
|
|
in_channels=block_in,
|
|
out_channels=block_out,
|
|
temb_channels=self.temb_ch,
|
|
dropout=dropout,
|
|
use_3d_conv=use_3d_conv,
|
|
half_3d=half_3d,
|
|
causal=causal,
|
|
)
|
|
)
|
|
block_in = block_out
|
|
if curr_res in attn_resolutions:
|
|
attn.append(make_attn_cls(block_in, attn_type=attn_type))
|
|
up = nn.Module()
|
|
up.block = block
|
|
up.attn = attn
|
|
if i_level != 0:
|
|
if half_t_mult:
|
|
up_time = i_level % 2 == 1
|
|
else:
|
|
up_time = True
|
|
up.upsample = Upsample3D(block_in, resamp_with_conv, up_time)
|
|
curr_res = curr_res * 2
|
|
self.up.insert(0, up) # prepend to get consistent order
|
|
|
|
# end
|
|
self.norm_out = Normalize(block_in)
|
|
self.conv_out = make_conv_cls(
|
|
block_in, out_ch, kernel_size=3, stride=1, padding=1
|
|
)
|
|
|
|
def _make_attn(self) -> Callable:
|
|
return make_attn
|
|
|
|
def _make_resblock(self) -> Callable:
|
|
return ResnetBlock3D
|
|
|
|
def _make_conv(self) -> Callable:
|
|
conv_cls = CausalConv3d if self.causal else nn.Conv3d
|
|
conv_cls = conv_cls if self.use_3d_conv else Conv2dWithExtraDim
|
|
return conv_cls
|
|
|
|
def get_last_layer(self, **kwargs):
|
|
return self.conv_out.weight
|
|
|
|
def forward(self, z, **kwargs):
|
|
# assert z.shape[1:] == self.z_shape[1:]
|
|
self.last_z_shape = z.shape
|
|
|
|
# timestep embedding
|
|
temb = None
|
|
|
|
# z to block_in
|
|
h = self.conv_in(z)
|
|
|
|
# middle
|
|
h = self.mid.block_1(h, temb, **kwargs)
|
|
h = self.mid.attn_1(h, **kwargs)
|
|
h = self.mid.block_2(h, temb, **kwargs)
|
|
|
|
# upsampling
|
|
for i_level in reversed(range(self.num_resolutions)):
|
|
for i_block in range(self.num_res_blocks + 1):
|
|
if self.gradient_checkpointing and self.training:
|
|
h = torch.utils.checkpoint.checkpoint(
|
|
self.up[i_level].block[i_block].__call__,
|
|
h,
|
|
temb,
|
|
use_reentrant=False,
|
|
**kwargs,
|
|
)
|
|
else:
|
|
h = self.up[i_level].block[i_block](h, temb, **kwargs)
|
|
if len(self.up[i_level].attn) > 0:
|
|
h = self.up[i_level].attn[i_block](h, **kwargs)
|
|
if i_level != 0:
|
|
h = self.up[i_level].upsample(h)
|
|
|
|
# end
|
|
if self.give_pre_end:
|
|
return h
|
|
|
|
h = self.norm_out(h)
|
|
h = nonlinearity(h)
|
|
h = self.conv_out(h, **kwargs)
|
|
if self.tanh_out:
|
|
h = torch.tanh(h)
|
|
return h
|