Spaces:
Running
on
Zero
Running
on
Zero
File size: 9,002 Bytes
323d67d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 |
from typing import Callable, List, Optional, Tuple, Union
from diffusers.models.attention_processor import Attention
from diffusers.models.embeddings import (
ImageProjection,
Resampler,
)
import torch
import torch.nn as nn
import torch.nn.functional as F
import copy
class IPAdapterAttnProcessor2_0(torch.nn.Module):
r"""
Attention processor for IP-Adapter for PyTorch 2.0.
Args:
hidden_size (`int`):
The hidden size of the attention layer.
cross_attention_dim (`int`):
The number of channels in the `encoder_hidden_states`.
num_tokens (`int`, `Tuple[int]` or `List[int]`, defaults to `(4,)`):
The context length of the image features.
scale (`float` or `List[float]`, defaults to 1.0):
the weight scale of image prompt.
"""
def __init__(
self, hidden_size, cross_attention_dim=None, num_tokens=(4,), scale=1.0
):
super().__init__()
if not hasattr(F, "scaled_dot_product_attention"):
raise ImportError(
f"{self.__class__.__name__} requires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0."
)
self.hidden_size = hidden_size
self.cross_attention_dim = cross_attention_dim
if not isinstance(num_tokens, (tuple, list)):
num_tokens = [num_tokens]
self.num_tokens = num_tokens
if not isinstance(scale, list):
scale = [scale] * len(num_tokens)
if len(scale) != len(num_tokens):
raise ValueError(
"`scale` should be a list of integers with the same length as `num_tokens`."
)
self.scale = scale
self.to_q_ip = nn.Linear(hidden_size, hidden_size, bias=False)
self.to_k_ip = nn.Linear(cross_attention_dim, hidden_size, bias=False)
self.to_v_ip = nn.Linear(cross_attention_dim, hidden_size, bias=False)
def __call__(
self,
attn: Attention,
hidden_states: torch.Tensor,
encoder_hidden_states: Optional[torch.Tensor] = None,
attention_mask: Optional[torch.Tensor] = None,
temb: Optional[torch.Tensor] = None,
scale: float = 1.0,
ip_adapter_masks: Optional[torch.Tensor] = None,
):
residual = hidden_states
# separate ip_hidden_states from encoder_hidden_states
if encoder_hidden_states is not None:
if isinstance(encoder_hidden_states, tuple):
encoder_hidden_states, ip_hidden_states = encoder_hidden_states
ip_hidden_states = ip_hidden_states[0]
if attn.spatial_norm is not None:
hidden_states = attn.spatial_norm(hidden_states, temb)
input_ndim = hidden_states.ndim
if input_ndim == 4:
batch_size, channel, height, width = hidden_states.shape
hidden_states = hidden_states.view(
batch_size, channel, height * width
).transpose(1, 2)
batch_size, sequence_length, _ = (
hidden_states.shape
if encoder_hidden_states is None
else encoder_hidden_states.shape
)
if attention_mask is not None:
attention_mask = attn.prepare_attention_mask(
attention_mask, sequence_length, batch_size
)
# scaled_dot_product_attention expects attention_mask shape to be
# (batch, heads, source_length, target_length)
attention_mask = attention_mask.view(
batch_size, attn.heads, -1, attention_mask.shape[-1]
)
if attn.group_norm is not None:
hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(
1, 2
)
query = attn.to_q(hidden_states)
if encoder_hidden_states is None:
encoder_hidden_states = hidden_states
elif attn.norm_cross:
encoder_hidden_states = attn.norm_encoder_hidden_states(
encoder_hidden_states
)
key = attn.to_k(encoder_hidden_states)
value = attn.to_v(encoder_hidden_states)
inner_dim = key.shape[-1]
head_dim = inner_dim // attn.heads
query = query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
key = key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
value = value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
# the output of sdp = (batch, num_heads, seq_len, head_dim)
# TODO: add support for attn.scale when we move to Torch 2.1
hidden_states = F.scaled_dot_product_attention(
query, key, value, attn_mask=attention_mask, dropout_p=0.0, is_causal=False
)
hidden_states = hidden_states.transpose(1, 2).reshape(
batch_size, -1, attn.heads * head_dim
)
hidden_states = hidden_states.to(query.dtype)
ip_query = self.to_q_ip(hidden_states)
ip_key = self.to_k_ip(ip_hidden_states)
ip_value = self.to_v_ip(ip_hidden_states)
ip_query = ip_query.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
ip_key = ip_key.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
ip_value = ip_value.view(batch_size, -1, attn.heads, head_dim).transpose(1, 2)
# the output of sdp = (batch, num_heads, seq_len, head_dim)
# TODO: add support for attn.scale when we move to Torch 2.1
current_ip_hidden_states = F.scaled_dot_product_attention(
ip_query,
ip_key,
ip_value,
attn_mask=None,
dropout_p=0.0,
is_causal=False,
)
current_ip_hidden_states = current_ip_hidden_states.transpose(1, 2).reshape(
batch_size, -1, attn.heads * head_dim
)
current_ip_hidden_states = current_ip_hidden_states.to(query.dtype)
hidden_states = hidden_states + scale * current_ip_hidden_states
# linear proj
hidden_states = attn.to_out[0](hidden_states)
# dropout
hidden_states = attn.to_out[1](hidden_states)
if input_ndim == 4:
hidden_states = hidden_states.transpose(-1, -2).reshape(
batch_size, channel, height, width
)
if attn.residual_connection:
hidden_states = hidden_states + residual
hidden_states = hidden_states / attn.rescale_output_factor
return hidden_states
def save_ip_adapter(unet, path):
state_dict = {}
if (
hasattr(unet, "encoder_hid_proj")
and unet.encoder_hid_proj is not None
and isinstance(unet.encoder_hid_proj, torch.nn.Module)
):
state_dict["encoder_hid_proj"] = unet.encoder_hid_proj.state_dict()
for name, module in unet.attn_processors.items():
if isinstance(module, torch.nn.Module):
state_dict[name] = module.state_dict()
torch.save(state_dict, path)
def load_ip_adapter(
unet,
path=None,
clip_embeddings_dim=1280,
cross_attention_dim=2048,
num_image_text_embeds=4,
):
if path is None:
state_dict = None
else:
state_dict = torch.load(path, map_location="cpu")
clip_embeddings_dim = state_dict["encoder_hid_proj"][
"image_projection_layers.0.image_embeds.weight"
].shape[-1]
num_image_text_embeds = (
state_dict["encoder_hid_proj"][
"image_projection_layers.0.image_embeds.weight"
].shape[0]
// cross_attention_dim
)
if not hasattr(unet, "encoder_hid_proj") or unet.encoder_hid_proj is None:
unet.encoder_hid_proj = MultiIPAdapterImageProjection(
[
ImageProjection(
cross_attention_dim=cross_attention_dim,
image_embed_dim=clip_embeddings_dim,
num_image_text_embeds=num_image_text_embeds,
)
]
).to(unet.device, unet.dtype)
if state_dict is not None:
unet.encoder_hid_proj.load_state_dict(state_dict["encoder_hid_proj"])
unet.config.encoder_hid_dim_type = "ip_image_proj"
for name, module in unet.named_modules():
if "attn2" in name and isinstance(module, Attention):
if not isinstance(module.processor, IPAdapterAttnProcessor2_0):
module.set_processor(
IPAdapterAttnProcessor2_0(
hidden_size=module.query_dim,
cross_attention_dim=cross_attention_dim,
scale=1.0,
).to(unet.device, unet.dtype)
)
if state_dict is not None:
module.processor.load_state_dict(state_dict[f"{name}.processor"])
def set_ip_adapter_scale(unet, scale=1.0):
for name, module in unet.named_modules():
if isinstance(module, IPAdapterAttnProcessor2_0):
module.scale = scale
|