mirror of
https://github.com/hpcaitech/ColossalAI.git
synced 2025-09-01 09:07:51 +00:00
support stable diffusion v2
This commit is contained in:
Fazzie
@@ -1,15 +1,11 @@
|
||||
import types
|
||||
|
||||
import torch
|
||||
import torch.nn as nn
|
||||
from functools import partial
|
||||
import clip
|
||||
from einops import rearrange, repeat
|
||||
from transformers import CLIPTokenizer, CLIPTextModel, CLIPTextConfig
|
||||
import kornia
|
||||
from transformers.models.clip.modeling_clip import CLIPTextTransformer
|
||||
from torch.utils.checkpoint import checkpoint
|
||||
|
||||
from ldm.modules.x_transformer import Encoder, TransformerWrapper # TODO: can we directly rely on lucidrains code and simply add this as a reuirement? --> test
|
||||
from transformers import T5Tokenizer, T5EncoderModel, CLIPTokenizer, CLIPTextModel
|
||||
|
||||
import open_clip
|
||||
from ldm.util import default, count_params
|
||||
|
||||
|
||||
class AbstractEncoder(nn.Module):
|
||||
@@ -20,169 +16,66 @@ class AbstractEncoder(nn.Module):
|
||||
raise NotImplementedError
|
||||
|
||||
|
||||
class IdentityEncoder(AbstractEncoder):
|
||||
|
||||
def encode(self, x):
|
||||
return x
|
||||
|
||||
|
||||
class ClassEmbedder(nn.Module):
|
||||
def __init__(self, embed_dim, n_classes=1000, key='class'):
|
||||
def __init__(self, embed_dim, n_classes=1000, key='class', ucg_rate=0.1):
|
||||
super().__init__()
|
||||
self.key = key
|
||||
self.embedding = nn.Embedding(n_classes, embed_dim)
|
||||
self.n_classes = n_classes
|
||||
self.ucg_rate = ucg_rate
|
||||
|
||||
def forward(self, batch, key=None):
|
||||
def forward(self, batch, key=None, disable_dropout=False):
|
||||
if key is None:
|
||||
key = self.key
|
||||
# this is for use in crossattn
|
||||
c = batch[key][:, None]
|
||||
if self.ucg_rate > 0. and not disable_dropout:
|
||||
mask = 1. - torch.bernoulli(torch.ones_like(c) * self.ucg_rate)
|
||||
c = mask * c + (1-mask) * torch.ones_like(c)*(self.n_classes-1)
|
||||
c = c.long()
|
||||
c = self.embedding(c)
|
||||
return c
|
||||
|
||||
def get_unconditional_conditioning(self, bs, device="cuda"):
|
||||
uc_class = self.n_classes - 1 # 1000 classes --> 0 ... 999, one extra class for ucg (class 1000)
|
||||
uc = torch.ones((bs,), device=device) * uc_class
|
||||
uc = {self.key: uc}
|
||||
return uc
|
||||
|
||||
class TransformerEmbedder(AbstractEncoder):
|
||||
"""Some transformer encoder layers"""
|
||||
def __init__(self, n_embed, n_layer, vocab_size, max_seq_len=77, device="cuda"):
|
||||
|
||||
def disabled_train(self, mode=True):
|
||||
"""Overwrite model.train with this function to make sure train/eval mode
|
||||
does not change anymore."""
|
||||
return self
|
||||
|
||||
|
||||
class FrozenT5Embedder(AbstractEncoder):
|
||||
"""Uses the T5 transformer encoder for text"""
|
||||
def __init__(self, version="google/t5-v1_1-large", device="cuda", max_length=77, freeze=True): # others are google/t5-v1_1-xl and google/t5-v1_1-xxl
|
||||
super().__init__()
|
||||
self.tokenizer = T5Tokenizer.from_pretrained(version)
|
||||
self.transformer = T5EncoderModel.from_pretrained(version)
|
||||
self.device = device
|
||||
self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
|
||||
attn_layers=Encoder(dim=n_embed, depth=n_layer))
|
||||
|
||||
def forward(self, tokens):
|
||||
tokens = tokens.to(self.device) # meh
|
||||
z = self.transformer(tokens, return_embeddings=True)
|
||||
return z
|
||||
|
||||
def encode(self, x):
|
||||
return self(x)
|
||||
|
||||
|
||||
class BERTTokenizer(AbstractEncoder):
|
||||
""" Uses a pretrained BERT tokenizer by huggingface. Vocab size: 30522 (?)"""
|
||||
def __init__(self, device="cuda", vq_interface=True, max_length=77):
|
||||
super().__init__()
|
||||
from transformers import BertTokenizerFast # TODO: add to reuquirements
|
||||
self.tokenizer = BertTokenizerFast.from_pretrained("bert-base-uncased")
|
||||
self.device = device
|
||||
self.vq_interface = vq_interface
|
||||
self.max_length = max_length
|
||||
|
||||
def forward(self, text):
|
||||
batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
|
||||
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
|
||||
tokens = batch_encoding["input_ids"].to(self.device)
|
||||
return tokens
|
||||
|
||||
@torch.no_grad()
|
||||
def encode(self, text):
|
||||
tokens = self(text)
|
||||
if not self.vq_interface:
|
||||
return tokens
|
||||
return None, None, [None, None, tokens]
|
||||
|
||||
def decode(self, text):
|
||||
return text
|
||||
|
||||
|
||||
class BERTEmbedder(AbstractEncoder):
|
||||
"""Uses the BERT tokenizr model and add some transformer encoder layers"""
|
||||
def __init__(self, n_embed, n_layer, vocab_size=30522, max_seq_len=77,
|
||||
device="cuda",use_tokenizer=True, embedding_dropout=0.0):
|
||||
super().__init__()
|
||||
self.use_tknz_fn = use_tokenizer
|
||||
if self.use_tknz_fn:
|
||||
self.tknz_fn = BERTTokenizer(vq_interface=False, max_length=max_seq_len)
|
||||
self.device = device
|
||||
self.transformer = TransformerWrapper(num_tokens=vocab_size, max_seq_len=max_seq_len,
|
||||
attn_layers=Encoder(dim=n_embed, depth=n_layer),
|
||||
emb_dropout=embedding_dropout)
|
||||
|
||||
def forward(self, text):
|
||||
if self.use_tknz_fn:
|
||||
tokens = self.tknz_fn(text)#.to(self.device)
|
||||
else:
|
||||
tokens = text
|
||||
z = self.transformer(tokens, return_embeddings=True)
|
||||
return z
|
||||
|
||||
def encode(self, text):
|
||||
# output of length 77
|
||||
return self(text)
|
||||
|
||||
|
||||
class SpatialRescaler(nn.Module):
|
||||
def __init__(self,
|
||||
n_stages=1,
|
||||
method='bilinear',
|
||||
multiplier=0.5,
|
||||
in_channels=3,
|
||||
out_channels=None,
|
||||
bias=False):
|
||||
super().__init__()
|
||||
self.n_stages = n_stages
|
||||
assert self.n_stages >= 0
|
||||
assert method in ['nearest','linear','bilinear','trilinear','bicubic','area']
|
||||
self.multiplier = multiplier
|
||||
self.interpolator = partial(torch.nn.functional.interpolate, mode=method)
|
||||
self.remap_output = out_channels is not None
|
||||
if self.remap_output:
|
||||
print(f'Spatial Rescaler mapping from {in_channels} to {out_channels} channels after resizing.')
|
||||
self.channel_mapper = nn.Conv2d(in_channels,out_channels,1,bias=bias)
|
||||
|
||||
def forward(self,x):
|
||||
for stage in range(self.n_stages):
|
||||
x = self.interpolator(x, scale_factor=self.multiplier)
|
||||
|
||||
|
||||
if self.remap_output:
|
||||
x = self.channel_mapper(x)
|
||||
return x
|
||||
|
||||
def encode(self, x):
|
||||
return self(x)
|
||||
|
||||
|
||||
class CLIPTextModelZero(CLIPTextModel):
|
||||
config_class = CLIPTextConfig
|
||||
|
||||
def __init__(self, config: CLIPTextConfig):
|
||||
super().__init__(config)
|
||||
self.text_model = CLIPTextTransformerZero(config)
|
||||
|
||||
class CLIPTextTransformerZero(CLIPTextTransformer):
|
||||
def _build_causal_attention_mask(self, bsz, seq_len):
|
||||
# lazily create causal attention mask, with full attention between the vision tokens
|
||||
# pytorch uses additive attention mask; fill with -inf
|
||||
mask = torch.empty(bsz, seq_len, seq_len)
|
||||
mask.fill_(float("-inf"))
|
||||
mask.triu_(1) # zero out the lower diagonal
|
||||
mask = mask.unsqueeze(1) # expand mask
|
||||
return mask.half()
|
||||
|
||||
class FrozenCLIPEmbedder(AbstractEncoder):
|
||||
"""Uses the CLIP transformer encoder for text (from Hugging Face)"""
|
||||
def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77, use_fp16=True):
|
||||
super().__init__()
|
||||
self.tokenizer = CLIPTokenizer.from_pretrained(version)
|
||||
|
||||
if use_fp16:
|
||||
self.transformer = CLIPTextModelZero.from_pretrained(version)
|
||||
else:
|
||||
self.transformer = CLIPTextModel.from_pretrained(version)
|
||||
|
||||
# print(self.transformer.modules())
|
||||
# print("check model dtyoe: {}, {}".format(self.tokenizer.dtype, self.transformer.dtype))
|
||||
self.device = device
|
||||
self.max_length = max_length
|
||||
self.freeze()
|
||||
self.max_length = max_length # TODO: typical value?
|
||||
if freeze:
|
||||
self.freeze()
|
||||
|
||||
def freeze(self):
|
||||
self.transformer = self.transformer.eval()
|
||||
#self.train = disabled_train
|
||||
for param in self.parameters():
|
||||
param.requires_grad = False
|
||||
|
||||
def forward(self, text):
|
||||
batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
|
||||
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
|
||||
# tokens = batch_encoding["input_ids"].to(self.device)
|
||||
tokens = batch_encoding["input_ids"].to(self.device)
|
||||
# print("token type: {}".format(tokens.dtype))
|
||||
outputs = self.transformer(input_ids=tokens)
|
||||
|
||||
z = outputs.last_hidden_state
|
||||
@@ -192,17 +85,80 @@ class FrozenCLIPEmbedder(AbstractEncoder):
|
||||
return self(text)
|
||||
|
||||
|
||||
class FrozenCLIPTextEmbedder(nn.Module):
|
||||
"""
|
||||
Uses the CLIP transformer encoder for text.
|
||||
"""
|
||||
def __init__(self, version='ViT-L/14', device="cuda", max_length=77, n_repeat=1, normalize=True):
|
||||
class FrozenCLIPEmbedder(AbstractEncoder):
|
||||
"""Uses the CLIP transformer encoder for text (from huggingface)"""
|
||||
LAYERS = [
|
||||
"last",
|
||||
"pooled",
|
||||
"hidden"
|
||||
]
|
||||
def __init__(self, version="openai/clip-vit-large-patch14", device="cuda", max_length=77,
|
||||
freeze=True, layer="last", layer_idx=None): # clip-vit-base-patch32
|
||||
super().__init__()
|
||||
self.model, _ = clip.load(version, jit=False, device="cpu")
|
||||
assert layer in self.LAYERS
|
||||
self.tokenizer = CLIPTokenizer.from_pretrained(version)
|
||||
self.transformer = CLIPTextModel.from_pretrained(version)
|
||||
self.device = device
|
||||
self.max_length = max_length
|
||||
self.n_repeat = n_repeat
|
||||
self.normalize = normalize
|
||||
if freeze:
|
||||
self.freeze()
|
||||
self.layer = layer
|
||||
self.layer_idx = layer_idx
|
||||
if layer == "hidden":
|
||||
assert layer_idx is not None
|
||||
assert 0 <= abs(layer_idx) <= 12
|
||||
|
||||
def freeze(self):
|
||||
self.transformer = self.transformer.eval()
|
||||
#self.train = disabled_train
|
||||
for param in self.parameters():
|
||||
param.requires_grad = False
|
||||
|
||||
def forward(self, text):
|
||||
batch_encoding = self.tokenizer(text, truncation=True, max_length=self.max_length, return_length=True,
|
||||
return_overflowing_tokens=False, padding="max_length", return_tensors="pt")
|
||||
tokens = batch_encoding["input_ids"].to(self.device)
|
||||
outputs = self.transformer(input_ids=tokens, output_hidden_states=self.layer=="hidden")
|
||||
if self.layer == "last":
|
||||
z = outputs.last_hidden_state
|
||||
elif self.layer == "pooled":
|
||||
z = outputs.pooler_output[:, None, :]
|
||||
else:
|
||||
z = outputs.hidden_states[self.layer_idx]
|
||||
return z
|
||||
|
||||
def encode(self, text):
|
||||
return self(text)
|
||||
|
||||
|
||||
class FrozenOpenCLIPEmbedder(AbstractEncoder):
|
||||
"""
|
||||
Uses the OpenCLIP transformer encoder for text
|
||||
"""
|
||||
LAYERS = [
|
||||
#"pooled",
|
||||
"last",
|
||||
"penultimate"
|
||||
]
|
||||
def __init__(self, arch="ViT-H-14", version="laion2b_s32b_b79k", device="cuda", max_length=77,
|
||||
freeze=True, layer="last"):
|
||||
super().__init__()
|
||||
assert layer in self.LAYERS
|
||||
model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), pretrained=version)
|
||||
del model.visual
|
||||
self.model = model
|
||||
|
||||
self.device = device
|
||||
self.max_length = max_length
|
||||
if freeze:
|
||||
self.freeze()
|
||||
self.layer = layer
|
||||
if self.layer == "last":
|
||||
self.layer_idx = 0
|
||||
elif self.layer == "penultimate":
|
||||
self.layer_idx = 1
|
||||
else:
|
||||
raise NotImplementedError()
|
||||
|
||||
def freeze(self):
|
||||
self.model = self.model.eval()
|
||||
@@ -210,55 +166,48 @@ class FrozenCLIPTextEmbedder(nn.Module):
|
||||
param.requires_grad = False
|
||||
|
||||
def forward(self, text):
|
||||
tokens = clip.tokenize(text).to(self.device)
|
||||
z = self.model.encode_text(tokens)
|
||||
if self.normalize:
|
||||
z = z / torch.linalg.norm(z, dim=1, keepdim=True)
|
||||
tokens = open_clip.tokenize(text)
|
||||
z = self.encode_with_transformer(tokens.to(self.device))
|
||||
return z
|
||||
|
||||
def encode(self, text):
|
||||
z = self(text)
|
||||
if z.ndim==2:
|
||||
z = z[:, None, :]
|
||||
z = repeat(z, 'b 1 d -> b k d', k=self.n_repeat)
|
||||
return z
|
||||
|
||||
|
||||
class FrozenClipImageEmbedder(nn.Module):
|
||||
"""
|
||||
Uses the CLIP image encoder.
|
||||
"""
|
||||
def __init__(
|
||||
self,
|
||||
model,
|
||||
jit=False,
|
||||
device='cuda' if torch.cuda.is_available() else 'cpu',
|
||||
antialias=False,
|
||||
):
|
||||
super().__init__()
|
||||
self.model, _ = clip.load(name=model, device=device, jit=jit)
|
||||
|
||||
self.antialias = antialias
|
||||
|
||||
self.register_buffer('mean', torch.Tensor([0.48145466, 0.4578275, 0.40821073]), persistent=False)
|
||||
self.register_buffer('std', torch.Tensor([0.26862954, 0.26130258, 0.27577711]), persistent=False)
|
||||
|
||||
def preprocess(self, x):
|
||||
# normalize to [0,1]
|
||||
x = kornia.geometry.resize(x, (224, 224),
|
||||
interpolation='bicubic',align_corners=True,
|
||||
antialias=self.antialias)
|
||||
x = (x + 1.) / 2.
|
||||
# renormalize according to clip
|
||||
x = kornia.enhance.normalize(x, self.mean, self.std)
|
||||
def encode_with_transformer(self, text):
|
||||
x = self.model.token_embedding(text) # [batch_size, n_ctx, d_model]
|
||||
x = x + self.model.positional_embedding
|
||||
x = x.permute(1, 0, 2) # NLD -> LND
|
||||
x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask)
|
||||
x = x.permute(1, 0, 2) # LND -> NLD
|
||||
x = self.model.ln_final(x)
|
||||
return x
|
||||
|
||||
def forward(self, x):
|
||||
# x is assumed to be in range [-1,1]
|
||||
return self.model.encode_image(self.preprocess(x))
|
||||
def text_transformer_forward(self, x: torch.Tensor, attn_mask = None):
|
||||
for i, r in enumerate(self.model.transformer.resblocks):
|
||||
if i == len(self.model.transformer.resblocks) - self.layer_idx:
|
||||
break
|
||||
if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting():
|
||||
x = checkpoint(r, x, attn_mask)
|
||||
else:
|
||||
x = r(x, attn_mask=attn_mask)
|
||||
return x
|
||||
|
||||
def encode(self, text):
|
||||
return self(text)
|
||||
|
||||
|
||||
class FrozenCLIPT5Encoder(AbstractEncoder):
|
||||
def __init__(self, clip_version="openai/clip-vit-large-patch14", t5_version="google/t5-v1_1-xl", device="cuda",
|
||||
clip_max_length=77, t5_max_length=77):
|
||||
super().__init__()
|
||||
self.clip_encoder = FrozenCLIPEmbedder(clip_version, device, max_length=clip_max_length)
|
||||
self.t5_encoder = FrozenT5Embedder(t5_version, device, max_length=t5_max_length)
|
||||
print(f"{self.clip_encoder.__class__.__name__} has {count_params(self.clip_encoder)*1.e-6:.2f} M parameters, "
|
||||
f"{self.t5_encoder.__class__.__name__} comes with {count_params(self.t5_encoder)*1.e-6:.2f} M params.")
|
||||
|
||||
def encode(self, text):
|
||||
return self(text)
|
||||
|
||||
def forward(self, text):
|
||||
clip_z = self.clip_encoder.encode(text)
|
||||
t5_z = self.t5_encoder.encode(text)
|
||||
return [clip_z, t5_z]
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
from ldm.util import count_params
|
||||
model = FrozenCLIPEmbedder()
|
||||
count_params(model, verbose=True)
|
||||
Reference in New Issue
Block a user