Spaces:
Running
on
Zero
Running
on
Zero
from dataclasses import dataclass | |
import torch | |
from torch import Tensor, nn | |
from flux.modules.layers import (DoubleStreamBlock, EmbedND, LastLayer, | |
MLPEmbedder, SingleStreamBlock, | |
SingleStreamBlock_kv,DoubleStreamBlock_kv, | |
timestep_embedding) | |
class FluxParams: | |
in_channels: int | |
vec_in_dim: int | |
context_in_dim: int | |
hidden_size: int | |
mlp_ratio: float | |
num_heads: int | |
depth: int | |
depth_single_blocks: int | |
axes_dim: list[int] | |
theta: int | |
qkv_bias: bool | |
guidance_embed: bool | |
class Flux(nn.Module): | |
""" | |
Transformer model for flow matching on sequences. | |
""" | |
def __init__(self, params: FluxParams,double_block_cls=DoubleStreamBlock,single_block_cls=SingleStreamBlock): | |
super().__init__() | |
self.params = params | |
self.in_channels = params.in_channels | |
self.out_channels = self.in_channels | |
if params.hidden_size % params.num_heads != 0: | |
raise ValueError( | |
f"Hidden size {params.hidden_size} must be divisible by num_heads {params.num_heads}" | |
) | |
pe_dim = params.hidden_size // params.num_heads | |
if sum(params.axes_dim) != pe_dim: | |
raise ValueError(f"Got {params.axes_dim} but expected positional dim {pe_dim}") | |
self.hidden_size = params.hidden_size | |
self.num_heads = params.num_heads | |
self.pe_embedder = EmbedND(dim=pe_dim, theta=params.theta, axes_dim=params.axes_dim) | |
self.img_in = nn.Linear(self.in_channels, self.hidden_size, bias=True) | |
self.time_in = MLPEmbedder(in_dim=256, hidden_dim=self.hidden_size) | |
self.vector_in = MLPEmbedder(params.vec_in_dim, self.hidden_size) | |
self.guidance_in = ( | |
MLPEmbedder(in_dim=256, hidden_dim=self.hidden_size) if params.guidance_embed else nn.Identity() | |
) | |
self.txt_in = nn.Linear(params.context_in_dim, self.hidden_size) | |
self.double_blocks = nn.ModuleList( | |
[ | |
double_block_cls( | |
self.hidden_size, | |
self.num_heads, | |
mlp_ratio=params.mlp_ratio, | |
qkv_bias=params.qkv_bias, | |
) | |
for _ in range(params.depth) | |
] | |
) | |
self.single_blocks = nn.ModuleList( | |
[ | |
single_block_cls(self.hidden_size, self.num_heads, mlp_ratio=params.mlp_ratio) | |
for _ in range(params.depth_single_blocks) | |
] | |
) | |
self.final_layer = LastLayer(self.hidden_size, 1, self.out_channels) | |
def forward( | |
self, | |
img: Tensor, | |
img_ids: Tensor, | |
txt: Tensor, | |
txt_ids: Tensor, | |
timesteps: Tensor, | |
y: Tensor, | |
guidance: Tensor | None = None, | |
) -> Tensor: | |
if img.ndim != 3 or txt.ndim != 3: | |
raise ValueError("Input img and txt tensors must have 3 dimensions.") | |
# running on sequences img | |
img = self.img_in(img) | |
vec = self.time_in(timestep_embedding(timesteps, 256)) | |
if self.params.guidance_embed: | |
if guidance is None: | |
raise ValueError("Didn't get guidance strength for guidance distilled model.") | |
vec = vec + self.guidance_in(timestep_embedding(guidance, 256)) | |
vec = vec + self.vector_in(y) | |
txt = self.txt_in(txt) | |
ids = torch.cat((txt_ids, img_ids), dim=1) | |
pe = self.pe_embedder(ids) | |
for block in self.double_blocks: | |
img, txt = block(img=img, txt=txt, vec=vec, pe=pe) | |
img = torch.cat((txt, img), 1) | |
for block in self.single_blocks: | |
img = block(img, vec=vec, pe=pe) | |
img = img[:, txt.shape[1] :, ...] | |
img = self.final_layer(img, vec) # (N, T, patch_size ** 2 * out_channels) | |
return img | |
class Flux_kv(Flux): | |
""" | |
继承Flux类,重写forward方法 | |
""" | |
def __init__(self, params: FluxParams,double_block_cls=DoubleStreamBlock_kv,single_block_cls=SingleStreamBlock_kv): | |
super().__init__(params,double_block_cls,single_block_cls) | |
def forward( | |
self, | |
img: Tensor, # (B,x,x) (1,4080,64) | |
img_ids: Tensor, | |
txt: Tensor, # torch.Size([1, 512, 4096]) | |
txt_ids: Tensor, | |
timesteps: Tensor, # torch.Size([1]) | |
y: Tensor, # torch.Size([1, 768]) | |
guidance: Tensor | None = None, # torch.Size([1]) | |
info: dict = {}, | |
) -> Tensor: | |
if img.ndim != 3 or txt.ndim != 3: | |
raise ValueError("Input img and txt tensors must have 3 dimensions.") | |
# running on sequences img | |
img = self.img_in(img) | |
vec = self.time_in(timestep_embedding(timesteps, 256)) # torch.Size([1, 3072]) | |
if self.params.guidance_embed: | |
if guidance is None: | |
raise ValueError("Didn't get guidance strength for guidance distilled model.") | |
vec = vec + self.guidance_in(timestep_embedding(guidance, 256)) # torch.Size([1, 3072]) | |
vec = vec + self.vector_in(y)# torch.Size([1, 3072]) | |
txt = self.txt_in(txt) # ([1, 512, 4096]) -> torch.Size([1, 512, 3072]) | |
ids = torch.cat((txt_ids, img_ids), dim=1) # torch.Size([1, 512, 3]) torch.Size([1, 4080, 3]) -> torch.Size([1, 4592, 3]) | |
pe = self.pe_embedder(ids) # torch.Size([1, 1, 4592, 64, 2, 2]) | |
if not info['inverse']: | |
mask_indices = info['mask_indices'] # 图片seq坐标下的 | |
info['pe_mask'] = torch.cat((pe[:, :, :512, ...],pe[:, :, mask_indices+512, ...]),dim=2) | |
cnt = 0 | |
for block in self.double_blocks: | |
info['id'] = cnt | |
img, txt = block(img=img, txt=txt, vec=vec, pe=pe, info=info) | |
cnt += 1 | |
cnt = 0 | |
x = torch.cat((txt, img), 1) | |
for block in self.single_blocks: | |
info['id'] = cnt | |
x = block(x, vec=vec, pe=pe, info=info) | |
cnt += 1 | |
img = x[:, txt.shape[1] :, ...] | |
img = self.final_layer(img, vec) # (N, T, patch_size ** 2 * out_channels) | |
return img | |