diffcve-2607 / diffcve /modules /attention.py
multimodalart's picture
multimodalart HF Staff
Upload diffcve/modules/attention.py with huggingface_hub
70619c6 verified
Raw History Blame Contribute Delete
15.2 kB
from inspect import isfunction
import torch
import torch.nn.functional as F
from torch import nn, einsum
from einops import rearrange, repeat
from diffcve.modules.diffusion.utils import checkpoint
from diffcve.modules.diffusion.utils import from_4d_to_3d, from_3d_to_4d
try:
import xformers
import xformers.ops
XFORMERS_AVAILABLE = True
except:
XFORMERS_AVAILABLE = False
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 max_neg_value(t):
return -torch.finfo(t.dtype).max
class GEGLU(nn.Module):
def __init__(self, dim_in, dim_out):
super().__init__()
self.proj = nn.Linear(dim_in, dim_out * 2)
def forward(self, x):
x, gate = self.proj(x).chunk(2, dim=-1)
return x * F.gelu(gate)
class FeedForward(nn.Module):
def __init__(self, dim, dim_out=None, mult=4, glu=False, dropout=0.):
super().__init__()
inner_dim = int(dim * mult)
dim_out = default(dim_out, dim)
project_in = nn.Sequential(
nn.Linear(dim, inner_dim),
nn.GELU()
) if not glu else GEGLU(dim, inner_dim)
self.net = nn.Sequential(
project_in,
nn.Dropout(dropout),
nn.Linear(inner_dim, dim_out)
)
def forward(self, x):
return self.net(x)
def zero_module(module):
for p in module.parameters():
p.detach().zero_()
return module
def Normalize(in_channels):
return torch.nn.GroupNorm(num_groups=32, num_channels=in_channels, eps=1e-6, affine=True)
class TemporalAttention(nn.Module):
def __init__(self, num_feat, num_heads=8, dim_head=64, num_frames=1):
super().__init__()
self.num_frames = num_frames
self.temporal_attn = MemoryEfficientSelfAttention(num_feat, heads=num_heads, dim_head=dim_head, dropout=0.0)
self.norm = nn.LayerNorm(num_feat)
self.temporal_alpha = nn.Parameter(torch.Tensor(1))
def forward(self, inp, t=None):
bt, c, h, w = inp.shape
# einops >= 0.4 raises when an explicit axis length is None (the authors'
# einops 0.3.0 inferred it from the shape); use the effective frame count.
t = t or self.num_frames
b = bt // t
ori = inp
inp = from_4d_to_3d(inp, b, c, t, h, w)
res = self.temporal_attn(self.norm(inp))
res = from_3d_to_4d(res, b, c, t, h, w)
out = self.temporal_alpha * res + (1 - self.temporal_alpha) * ori
return out
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 CrossAttention(nn.Module):
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.):
super().__init__()
inner_dim = dim_head * heads
context_dim = default(context_dim, query_dim)
self.scale = dim_head ** -0.5
self.heads = heads
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)
)
def forward(self, x, context=None, mask=None):
h = self.heads
q = self.to_q(x)
context = default(context, x)
k = self.to_k(context)
v = self.to_v(context)
q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
if exists(mask):
mask = rearrange(mask, 'b ... -> b (...)')
max_neg_value = -torch.finfo(sim.dtype).max
mask = repeat(mask, 'b j -> (b h) () j', h=h)
sim.masked_fill_(~mask, max_neg_value)
attn = sim.softmax(dim=-1)
out = einsum('b i j, b j d -> b i d', attn, v)
out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
return self.to_out(out)
class MemoryEfficientSelfAttention(nn.Module):
# https://github.com/MatthieuTPHR/diffusers/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers/models/attention.py#L223
def __init__(self, query_dim, heads=8, dim_head=64, dropout=0.0):
super().__init__()
print(f"Setting up {self.__class__.__name__}. Query dim is {query_dim}, using {heads} heads.")
inner_dim = dim_head * heads
self.heads = heads
self.dim_head = dim_head
self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
self.to_k = nn.Linear(query_dim, inner_dim, bias=False)
self.to_v = nn.Linear(query_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, mask=None):
q = self.to_q(x)
k = self.to_k(x)
v = self.to_v(x)
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),
)
if torch.onnx.is_in_onnx_export():
scale = self.dim_head ** -0.5
attention = torch.matmul(q, k.transpose(-1, -2)) * scale
attention = attention.softmax(dim=-1)
out = torch.matmul(attention, v)
else:
out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=self.attention_op)
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)
)
return self.to_out(out)
class MemoryEfficientCrossAttention(nn.Module):
# https://github.com/MatthieuTPHR/diffusers/blob/d80b531ff8060ec1ea982b65a1b8df70f73aa67c/src/diffusers/models/attention.py#L223
def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.0):
super().__init__()
print(f"Setting up {self.__class__.__name__}. Query dim is {query_dim}, context_dim is {context_dim} and using "
f"{heads} heads.")
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):
q = self.to_q(x)
context = default(context, x)
if x.shape[0] != context.shape[0]:
context = torch.repeat_interleave(context, x.shape[0] // context.shape[0], dim=0)
k = self.to_k(context)
v = self.to_v(context)
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),
)
if torch.onnx.is_in_onnx_export():
scale = self.dim_head ** -0.5
attention = torch.matmul(q, k.transpose(-1, -2)) * scale
attention = attention.softmax(dim=-1)
out = torch.matmul(attention, v)
else:
out = xformers.ops.memory_efficient_attention(q, k, v, attn_bias=None, op=self.attention_op)
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)
)
return self.to_out(out)
class BasicTransformerBlock(nn.Module):
def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=False):
super().__init__()
self.attn1 = CrossAttention(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout)
self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
self.attn2 = CrossAttention(query_dim=dim, context_dim=context_dim,
heads=n_heads, dim_head=d_head, dropout=dropout)
self.norm1 = nn.LayerNorm(dim)
self.norm2 = nn.LayerNorm(dim)
self.norm3 = nn.LayerNorm(dim)
self.checkpoint = checkpoint
def forward(self, x, context=None):
return checkpoint(self._forward, (x, context), self.parameters(), self.checkpoint)
def _forward(self, x, context=None):
x = self.attn1(self.norm1(x)) + x
x = self.attn2(self.norm2(x), context=context) + x
x = self.ff(self.norm3(x)) + x
return x
class ConfigurableTransformerBlock(nn.Module):
ATTENTION_MODES = {
"softmax": CrossAttention,
"softmax-xformers": MemoryEfficientCrossAttention
}
def __init__(self, dim, n_heads, d_head, dropout=0., context_dim=None, gated_ff=True, checkpoint=True,
disable_self_attn=False):
super().__init__()
attn_mode = "softmax-xformers" if XFORMERS_AVAILABLE else "softmax"
assert attn_mode in self.ATTENTION_MODES
attn_cls = self.ATTENTION_MODES[attn_mode]
self.disable_self_attn = disable_self_attn
self.attn1 = attn_cls(query_dim=dim, heads=n_heads, dim_head=d_head, dropout=dropout,
context_dim=context_dim if self.disable_self_attn else None)
self.ff = FeedForward(dim, dropout=dropout, glu=gated_ff)
self.attn2 = attn_cls(query_dim=dim, context_dim=context_dim,
heads=n_heads, dim_head=d_head, dropout=dropout)
self.norm1 = nn.LayerNorm(dim)
self.norm2 = nn.LayerNorm(dim)
self.norm3 = nn.LayerNorm(dim)
self.checkpoint = checkpoint
def forward(self, x, context=None):
return checkpoint(self._forward, (x, context), self.parameters(), self.checkpoint)
def _forward(self, x, context=None):
x = self.attn1(self.norm1(x), context=context if self.disable_self_attn else None) + x
x = self.attn2(self.norm2(x), context=context) + x
x = self.ff(self.norm3(x)) + x
return x
class SpatialTransformer(nn.Module):
def __init__(self, in_channels, n_heads, d_head,
depth=1, dropout=0., context_dim=None):
super().__init__()
self.in_channels = in_channels
inner_dim = n_heads * d_head
self.norm = Normalize(in_channels)
self.proj_in = nn.Conv2d(in_channels,
inner_dim,
kernel_size=1,
stride=1,
padding=0)
self.transformer_blocks = nn.ModuleList(
[BasicTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim)
for d in range(depth)]
)
self.proj_out = zero_module(nn.Conv2d(inner_dim,
in_channels,
kernel_size=1,
stride=1,
padding=0))
def forward(self, x, context=None):
b, c, h, w = x.shape
x_in = x
x = self.norm(x)
x = self.proj_in(x)
x = rearrange(x, 'b c h w -> b (h w) c')
for block in self.transformer_blocks:
x = block(x, context=context)
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w)
x = self.proj_out(x)
return x + x_in
class ProjectedSpatialTransformer(nn.Module):
def __init__(self, in_channels, n_heads, d_head,
depth=1, dropout=0., context_dim=None,
disable_self_attn=False, use_linear=False,
use_checkpoint=False):
super().__init__()
if exists(context_dim) and not isinstance(context_dim, list):
context_dim = [context_dim]
self.in_channels = in_channels
inner_dim = n_heads * d_head
self.norm = Normalize(in_channels)
if not use_linear:
self.proj_in = nn.Conv2d(in_channels,
inner_dim,
kernel_size=1,
stride=1,
padding=0)
else:
self.proj_in = nn.Linear(in_channels, inner_dim)
self.transformer_blocks = nn.ModuleList(
[ConfigurableTransformerBlock(inner_dim, n_heads, d_head, dropout=dropout, context_dim=context_dim[d],
disable_self_attn=disable_self_attn, checkpoint=use_checkpoint)
for d in range(depth)]
)
if not use_linear:
self.proj_out = zero_module(nn.Conv2d(inner_dim,
in_channels,
kernel_size=1,
stride=1,
padding=0))
else:
self.proj_out = zero_module(nn.Linear(in_channels, inner_dim))
self.use_linear = use_linear
def forward(self, x, context=None):
if not isinstance(context, list):
context = [context]
b, c, h, w = x.shape
x_in = x
x = self.norm(x)
if not self.use_linear:
x = self.proj_in(x)
x = rearrange(x, 'b c h w -> b (h w) c').contiguous()
if self.use_linear:
x = self.proj_in(x)
for i, block in enumerate(self.transformer_blocks):
x = block(x, context=context[i])
if self.use_linear:
x = self.proj_out(x)
x = rearrange(x, 'b (h w) c -> b c h w', h=h, w=w).contiguous()
if not self.use_linear:
x = self.proj_out(x)
return x + x_in