UniTok / model /quant.py
machuofan
init
7385f22
from typing import List, Tuple
import torch
from torch import distributed as tdist, nn as nn
from torch.nn import functional as F
from torch.nn.functional import scaled_dot_product_attention
# from utils import dist
# this file only provides the VectorQuantizer2 used in VQVAE
__all__ = ['VectorQuantizer', ]
def get_entropy_loss(latent_embed, codebook_embed, inv_entropy_tau):
E_dist = latent_embed.square().sum(dim=1, keepdim=True) + codebook_embed.square().sum(dim=1, keepdim=False)
E_dist.addmm_(latent_embed, codebook_embed.T, alpha=-2, beta=1) # E_dist: (N, vocab_size)
logits = -E_dist.float().mul_(inv_entropy_tau)
# calc per_sample_entropy
prob, log_prob = logits.softmax(dim=-1), logits.log_softmax(dim=-1) # both are (N, vocab_size)
per_sample_entropy = torch.mean((-prob * log_prob).sum(dim=-1))
# calc codebook_entropy
avg_prob = prob.mean(dim=0) # (vocab_size,)
log_avg_prob = torch.log(avg_prob + 1e-7)
codebook_entropy = (-avg_prob * log_avg_prob).sum()
# calc entropy_loss
entropy_loss = per_sample_entropy - codebook_entropy
return entropy_loss
class NormalizedEmbedding(nn.Embedding):
def __init__(self, num_embeddings: int, embedding_dim: int):
super().__init__(num_embeddings=num_embeddings, embedding_dim=embedding_dim)
# self.norm_scale = nn.Parameter(torch.tensor(0.0, dtype=torch.float32))
def forward(self, idx):
return F.embedding(
idx, F.normalize(self.weight, dim=1), self.padding_idx, self.max_norm,
self.norm_type, self.scale_grad_by_freq, self.sparse
)
def get_norm_weight(self):
return F.normalize(self.weight, dim=1)
class ResConv(nn.Conv2d):
def __init__(self, embed_dim, quant_resi):
ks = 3 if quant_resi < 0 else 1
super().__init__(in_channels=embed_dim, out_channels=embed_dim, kernel_size=ks, stride=1, padding=ks // 2)
self.resi_ratio = abs(quant_resi)
def forward(self, h_BChw):
return h_BChw.mul(1 - self.resi_ratio) + super().forward(h_BChw).mul_(self.resi_ratio)
class VectorQuantizer(nn.Module):
def __init__(
self, vocab_size: int, vocab_width: int, vocab_norm: bool, beta: float = 0.25, quant_resi=-0.5,
using_entropy_loss=False, entropy_temp=0.01,
):
super().__init__()
self.vocab_size: int = vocab_size
self.vocab_width: int = vocab_width
self.register_buffer('vocab_usage', torch.zeros(self.vocab_size))
self.vocab_usage_record_times: int = 0
self.vocab_norm: bool = vocab_norm
# self.quant_resi = ResConv(self.vocab_width, quant_resi=quant_resi)
self.quant_resi = nn.Identity()
self.embedding = nn.Embedding(self.vocab_size, self.vocab_width)
self.beta: float = beta
self.using_entropy_loss, self.inv_entropy_tau = using_entropy_loss, 1 / entropy_temp
if not self.vocab_norm:
assert not self.using_entropy_loss, 'entropy loss without vocab norm is not supported'
def init_vocab(self, eini: float):
if eini > 0:
nn.init.trunc_normal_(self.embedding.weight.data, std=eini)
elif eini < 0:
base = self.vocab_width ** -0.5
base /= 36
self.embedding.weight.data.uniform_(-abs(eini) * base, abs(eini) * base)
def extra_repr(self) -> str:
return f'beta={self.beta:g}'
# ===================== `forward` is only used in VAE training =====================
def forward(self, f_BChw: torch.Tensor, ret_usages=False) -> Tuple[
torch.Tensor, torch.Tensor, torch.Tensor, List[float]]:
f_BChw = f_BChw.float()
B, C, h, w = f_BChw.shape
if self.vocab_norm:
if self.using_entropy_loss:
# find the nearest neighbor
NxC = f_BChw.permute(0, 2, 3, 1).reshape(-1, C)
NxC_no_grad = NxC.detach()
NxC_no_grad = F.normalize(NxC_no_grad, dim=-1)
idx_N = torch.argmax(NxC_no_grad @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
# get logits
E_dist = NxC.square().sum(dim=1, keepdim=True) + self.embedding.weight.square().sum(dim=1,
keepdim=False)
E_dist.addmm_(NxC, self.embedding.weight.T, alpha=-2, beta=1) # E_dist: (N, vocab_size)
logits = -E_dist.float().mul_(self.inv_entropy_tau)
# calc per_sample_entropy
prob, log_prob = logits.softmax(dim=-1), logits.log_softmax(dim=-1) # both are (N, vocab_size)
per_sample_entropy = torch.mean((-prob * log_prob).sum(dim=-1))
# calc codebook_entropy
avg_prob = prob.mean(dim=0) # (vocab_size,)
log_avg_prob = torch.log(avg_prob + 1e-7)
codebook_entropy = (-avg_prob * log_avg_prob).sum()
# calc entropy_loss
entropy_loss = per_sample_entropy - codebook_entropy
else:
NxC_no_grad = f_BChw.detach().permute(0, 2, 3, 1).reshape(-1, C)
NxC_no_grad = F.normalize(NxC_no_grad, dim=-1)
idx_N = torch.argmax(NxC_no_grad @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
entropy_loss = 0
else: # not self.vocab_norm
NxC_no_grad = f_BChw.detach().permute(0, 2, 3, 1).reshape(-1, C)
E_dist = NxC_no_grad.square().sum(dim=1, keepdim=True) + self.embedding.weight.data.square().sum(dim=1,
keepdim=False)
E_dist.addmm_(NxC_no_grad, self.embedding.weight.data.T, alpha=-2, beta=1) # E_dist: N x vocab_size
idx_N = torch.argmin(E_dist, dim=1)
entropy_loss = 0
prob_per_class_is_chosen = idx_N.bincount(minlength=self.vocab_size).float()
handler = tdist.all_reduce(prob_per_class_is_chosen, async_op=True) if (
self.training and dist.initialized()) else None
# look up
idx_Bhw = idx_N.view(B, h, w)
fhat_BChw = self.quant_resi(self.embedding(idx_Bhw).permute(0, 3, 1, 2).contiguous())
# calc loss
vq_loss = F.mse_loss(fhat_BChw.detach(), f_BChw).mul_(self.beta) + F.mse_loss(fhat_BChw, f_BChw.detach())
fhat_BChw = (fhat_BChw.detach() - f_BChw.detach()).add_(f_BChw)
# update vocab_usage
if handler is not None:
handler.wait()
prob_per_class_is_chosen /= prob_per_class_is_chosen.sum()
vocab_usage = (prob_per_class_is_chosen > 0.01 / self.vocab_size).float().mean().mul_(100)
if self.vocab_usage_record_times == 0:
self.vocab_usage.copy_(prob_per_class_is_chosen)
elif self.vocab_usage_record_times < 100:
self.vocab_usage.mul_(0.9).add_(prob_per_class_is_chosen, alpha=0.1)
else:
self.vocab_usage.mul_(0.99).add_(prob_per_class_is_chosen, alpha=0.01)
self.vocab_usage_record_times += 1
return fhat_BChw, vq_loss, entropy_loss, (vocab_usage if ret_usages else None)
def f_to_idx(self, f_BChw: torch.Tensor) -> torch.LongTensor:
f_BChw = f_BChw.float()
B, C, h, w = f_BChw.shape
with torch.cuda.amp.autocast(enabled=False):
# find the nearest embedding
query_NxC = f_BChw.detach().permute(0, 2, 3, 1).reshape(-1, C)
if self.vocab_norm:
query_NxC = F.normalize(query_NxC, dim=-1)
idx_N = torch.argmax(query_NxC @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
else:
E_dist = torch.sum(query_NxC.square(), dim=1, keepdim=True) + torch.sum(
self.embedding.weight.data.square(), dim=1, keepdim=False)
E_dist.addmm_(query_NxC, self.embedding.weight.data.T, alpha=-2, beta=1) # (B*h*w, vocab_size)
idx_N = torch.argmin(E_dist, dim=1)
return idx_N.view(B, h, w)
class VectorQuantizerHybrid(nn.Module):
def __init__(
self, vocab_size: int, vocab_width: int, vocab_norm: bool, beta: float = 0.25, quant_resi=-0.5,
using_entropy_loss=False, entropy_temp=0.01,
):
super().__init__()
self.vocab_size: int = vocab_size
self.vocab_width: int = vocab_width
self.register_buffer('vocab_usage', torch.zeros(self.vocab_size))
self.vocab_usage_record_times: int = 0
self.vocab_norm: bool = vocab_norm
# self.quant_resi = ResConv(self.vocab_width, quant_resi=quant_resi)
self.embedding = nn.Embedding(self.vocab_size, self.vocab_width)
self.beta: float = beta
self.using_entropy_loss, self.inv_entropy_tau = using_entropy_loss, 1 / entropy_temp
if not self.vocab_norm:
assert not self.using_entropy_loss, 'entropy loss without vocab norm is not supported'
def init_vocab(self, eini: float):
if eini > 0:
nn.init.trunc_normal_(self.embedding.weight.data, std=eini)
elif eini < 0:
base = self.vocab_width ** -0.5
base /= 36
self.embedding.weight.data.uniform_(-abs(eini) * base, abs(eini) * base)
def extra_repr(self) -> str:
return f'beta={self.beta:g}'
def forward(self, class_tokens, patch_tokens, ret_usages=False):
class_tokens = class_tokens.float()
patch_tokens = patch_tokens.float()
B, L, C = class_tokens.shape
B, C, H, W = patch_tokens.shape
patch_tokens = patch_tokens.flatten(start_dim=2).permute(0, 2, 1)
NxC = torch.cat((class_tokens, patch_tokens), dim=1).reshape(-1, C)
if self.vocab_norm:
if self.using_entropy_loss:
# find the nearest neighbor
NxC_no_grad = NxC.detach()
NxC_no_grad = F.normalize(NxC_no_grad, dim=-1)
idx_N = torch.argmax(NxC_no_grad @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
# get logits
E_dist = NxC.square().sum(dim=1, keepdim=True) + self.embedding.weight.square().sum(dim=1,
keepdim=False)
E_dist.addmm_(NxC, self.embedding.weight.T, alpha=-2, beta=1) # E_dist: (N, vocab_size)
logits = -E_dist.float().mul_(self.inv_entropy_tau)
# calc per_sample_entropy
prob, log_prob = logits.softmax(dim=-1), logits.log_softmax(dim=-1) # both are (N, vocab_size)
per_sample_entropy = torch.mean((-prob * log_prob).sum(dim=-1))
# calc codebook_entropy
avg_prob = prob.mean(dim=0) # (vocab_size,)
log_avg_prob = torch.log(avg_prob + 1e-7)
codebook_entropy = (-avg_prob * log_avg_prob).sum()
# calc entropy_loss
entropy_loss = per_sample_entropy - codebook_entropy
else:
NxC_no_grad = NxC.detach()
NxC_no_grad = F.normalize(NxC_no_grad, dim=-1)
idx_N = torch.argmax(NxC_no_grad @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
entropy_loss = 0
else: # not self.vocab_norm
NxC_no_grad = NxC.detach()
E_dist = NxC_no_grad.square().sum(dim=1, keepdim=True) + self.embedding.weight.data.square().sum(dim=1,
keepdim=False)
E_dist.addmm_(NxC_no_grad, self.embedding.weight.data.T, alpha=-2, beta=1) # E_dist: N x vocab_size
idx_N = torch.argmin(E_dist, dim=1)
entropy_loss = 0
prob_per_class_is_chosen = idx_N.bincount(minlength=self.vocab_size).float()
handler = tdist.all_reduce(prob_per_class_is_chosen, async_op=True) if (
self.training and dist.initialized()) else None
# look up
fhat = self.embedding(idx_N)
# calc loss
vq_loss = F.mse_loss(fhat.detach(), NxC).mul_(self.beta) + F.mse_loss(fhat, NxC.detach())
fhat = (fhat.detach() - NxC.detach()).add_(NxC)
# update vocab_usage
if handler is not None:
handler.wait()
prob_per_class_is_chosen /= prob_per_class_is_chosen.sum()
vocab_usage = (prob_per_class_is_chosen > 0.01 / self.vocab_size).float().mean().mul_(100)
if self.vocab_usage_record_times == 0:
self.vocab_usage.copy_(prob_per_class_is_chosen)
elif self.vocab_usage_record_times < 100:
self.vocab_usage.mul_(0.9).add_(prob_per_class_is_chosen, alpha=0.1)
else:
self.vocab_usage.mul_(0.99).add_(prob_per_class_is_chosen, alpha=0.01)
self.vocab_usage_record_times += 1
fhat = fhat.view(B, -1, C)
fhat_class = fhat[:, :L, :]
fhat_patch = fhat[:, L:, :].view(B, H, W, C).permute(0, 3, 1, 2)
return fhat_class, fhat_patch, vq_loss, entropy_loss, (vocab_usage if ret_usages else None)
def f_to_idx(self, class_tokens, patch_tokens) -> torch.LongTensor:
B, L, C = class_tokens.shape
B, C, H, W = patch_tokens.shape
class_tokens = class_tokens.float()
patch_tokens = patch_tokens.float()
patch_tokens = patch_tokens.flatten(start_dim=2).permute(0, 2, 1)
NxC = torch.cat((class_tokens, patch_tokens), dim=1).reshape(-1, C)
with torch.cuda.amp.autocast(enabled=False):
# find the nearest embedding
if self.vocab_norm:
NxC = F.normalize(NxC, dim=-1)
idx_N = torch.argmax(NxC @ F.normalize(self.embedding.weight.data.T, dim=0), dim=1)
else:
E_dist = torch.sum(NxC.square(), dim=1, keepdim=True) + torch.sum(self.embedding.weight.data.square(),
dim=1, keepdim=False)
E_dist.addmm_(NxC, self.embedding.weight.data.T, alpha=-2, beta=1) # (B*h*w, vocab_size)
idx_N = torch.argmin(E_dist, dim=1)
return idx_N
class VectorQuantizerX(nn.Module):
def __init__(
self,
vocab_size: int,
vocab_width: int,
beta: float = 0.25,
use_entropy_loss=False,
entropy_temp=0.01,
):
super().__init__()
self.beta = beta
self.vocab_size = vocab_size
self.vocab_width = vocab_width
self.vocab_usage_record_times: int = 0
self.register_buffer('vocab_usage', torch.zeros(self.vocab_size))
self.codebook = NormalizedEmbedding(self.vocab_size, self.vocab_width)
self.use_entropy_loss = use_entropy_loss
self.inv_entropy_tau = 1 / entropy_temp
def init_vocab(self, eini: float):
if eini > 0:
nn.init.trunc_normal_(self.codebook.weight.data, std=eini)
elif eini < 0:
base = self.vocab_width ** -0.5
base /= 36
self.codebook.weight.data.uniform_(-abs(eini) * base, abs(eini) * base)
def extra_repr(self) -> str:
return f'beta={self.beta:g}'
def forward(self, features):
B, L, C = features.shape
features = features.reshape(-1, C)
features = F.normalize(features, dim=-1).float()
codebook_embed = self.codebook.get_norm_weight()
indices = torch.argmax(features.detach() @ codebook_embed.T, dim=1)
entropy_loss = get_entropy_loss(features, codebook_embed, self.inv_entropy_tau) if self.use_entropy_loss else 0
features_hat = self.codebook(indices)
# calc loss
vq_loss = F.mse_loss(features_hat.detach(), features).mul_(self.beta) + F.mse_loss(features_hat,
features.detach())
features_hat = (features_hat.detach() - features.detach()).add_(features)
# update vocab_usage
prob_per_class_is_chosen = indices.bincount(minlength=self.vocab_size).float()
handler = tdist.all_reduce(prob_per_class_is_chosen, async_op=True) if (
self.training and dist.initialized()) else None
if handler is not None:
handler.wait()
prob_per_class_is_chosen /= prob_per_class_is_chosen.sum()
vocab_usage = (prob_per_class_is_chosen > 0.01 / self.vocab_size).float().mean().mul_(100)
if self.vocab_usage_record_times == 0:
self.vocab_usage.copy_(prob_per_class_is_chosen)
elif self.vocab_usage_record_times < 100:
self.vocab_usage.mul_(0.9).add_(prob_per_class_is_chosen, alpha=0.1)
else:
self.vocab_usage.mul_(0.99).add_(prob_per_class_is_chosen, alpha=0.01)
self.vocab_usage_record_times += 1
return features_hat.view(B, L, C), vq_loss, entropy_loss, vocab_usage
def f_to_idx(self, features):
B, L, C = features.shape
features = features.reshape(-1, C)
features = F.normalize(features, dim=-1).float()
codebook_embed = self.codebook.get_norm_weight().float()
indices = torch.argmax(features.detach() @ codebook_embed.T, dim=1)
return indices.view(B, L)
class VectorQuantizerM(nn.Module):
def __init__(
self,
vocab_size,
vocab_width,
beta=0.25,
use_entropy_loss=False,
entropy_temp=0.01,
num_codebooks=16
):
super().__init__()
self.num_codebooks = num_codebooks
self.codebooks = nn.ModuleList()
for _ in range(num_codebooks):
codebook = VectorQuantizerX(
vocab_size=vocab_size // num_codebooks,
vocab_width=vocab_width // num_codebooks,
beta=beta,
use_entropy_loss=use_entropy_loss,
entropy_temp=entropy_temp,
)
self.codebooks.append(codebook)
def init_vocab(self, eini: float):
for codebook in self.codebooks:
codebook.init_vocab(eini)
def f_to_idx(self, features):
indices = []
chunk_size = features.shape[-1] // self.num_codebooks
splited_features = features.split(chunk_size, dim=-1)
for i, codebook in enumerate(self.codebooks):
indices.append(codebook.f_to_idx(splited_features[i]))
indices = torch.stack(indices, dim=1)
return indices
def idx_to_f(self, indices):
assert indices.shape[1] == self.num_codebooks
latent_features = []
for i, codebook in enumerate(self.codebooks):
sub_indices = indices[:, i].flatten(start_dim=1)
latent_feature = codebook.codebook(sub_indices)
latent_features.append(latent_feature)
latent_features = torch.cat(latent_features, dim=-1)
return latent_features
def forward(self, features):
latent_features = []
global_vq_loss = 0.
global_entropy_loss = 0.
global_vocab_usage = 0.
chunk_size = features.shape[-1] // self.num_codebooks
splited_features = features.split(chunk_size, dim=-1)
for i, codebook in enumerate(self.codebooks):
latent_feature, vq_loss, entropy_loss, vocab_usage = codebook(splited_features[i])
latent_features.append(latent_feature)
global_vq_loss += vq_loss
global_entropy_loss += entropy_loss
global_vocab_usage += vocab_usage
latent_features = torch.cat(latent_features, dim=-1)
global_entropy_loss /= self.num_codebooks
global_vq_loss /= self.num_codebooks
global_vocab_usage /= self.num_codebooks
return latent_features, global_vq_loss, global_entropy_loss, global_vocab_usage
class CausalAttention(nn.Module):
def __init__(self, in_dim, out_dim, num_heads):
super().__init__()
if in_dim > out_dim:
# assert in_dim // num_heads == out_dim
self.head_dim = in_dim // num_heads
self.qkv = nn.Linear(in_dim, in_dim * 3, bias=False)
self.q_bias = nn.Parameter(torch.zeros(in_dim))
self.v_bias = nn.Parameter(torch.zeros(in_dim))
self.register_buffer('zero_k_bias', torch.zeros(in_dim))
else:
# assert out_dim // num_heads == in_dim
self.head_dim = out_dim // num_heads
self.qkv = nn.Linear(in_dim, out_dim * 3, bias=False)
self.q_bias = nn.Parameter(torch.zeros(out_dim))
self.v_bias = nn.Parameter(torch.zeros(out_dim))
self.register_buffer('zero_k_bias', torch.zeros(out_dim))
self.in_dim = in_dim
self.out_dim = out_dim
self.num_heads = num_heads
self.scale = self.head_dim ** -0.5
self.proj = nn.Linear(out_dim, out_dim)
def forward(self, x: torch.Tensor) -> torch.Tensor:
B, N, C = x.shape
qkv = F.linear(input=x, weight=self.qkv.weight, bias=torch.cat((self.q_bias, self.zero_k_bias, self.v_bias)))
q, k, v = qkv.reshape(B, N, 3, self.num_heads, self.head_dim).permute(2, 0, 3, 1, 4).unbind(0)
x = scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=0., is_causal=True)
if self.in_dim > self.out_dim:
x = torch.mean(x, dim=1)
if self.in_dim // self.num_heads != self.out_dim:
x = nn.functional.adaptive_avg_pool1d(x, self.out_dim)
else:
x = x.transpose(1, 2).reshape(B, N, -1)
x = self.proj(x)
return x
class AttnProjection(nn.Module):
def __init__(self, in_dim, out_dim, num_heads, norm_layer=nn.LayerNorm, mlp_ratio=2):
super().__init__()
assert out_dim % in_dim == 0 or in_dim % out_dim == 0
self.in_dim = in_dim
self.out_dim = out_dim
self.norm1 = norm_layer(in_dim)
self.attn = CausalAttention(in_dim, out_dim, num_heads)
self.proj = nn.Linear(in_dim, out_dim)
self.norm3 = norm_layer(in_dim)
self.norm2 = norm_layer(out_dim)
hidden_dim = int(out_dim * mlp_ratio)
self.mlp = GeGluMlp(
in_features=out_dim,
hidden_features=hidden_dim
)
def forward(self, x):
x = self.proj(self.norm3(x)) + self.attn(self.norm1(x))
x = x + self.mlp(self.norm2(x))
return x
from functools import partial
from timm.models.layers import create_conv2d, get_norm_act_layer, get_norm_layer, make_divisible
class GeGluMlp(nn.Module):
def __init__(
self,
in_features,
hidden_features,
act_layer = None,
drop = 0.0,
):
super().__init__()
norm_layer = partial(get_norm_layer('layernorm'), eps=1e-6)
self.norm = norm_layer(in_features)
self.act = nn.GELU(approximate='tanh')
self.w0 = nn.Linear(in_features, hidden_features)
self.w1 = nn.Linear(in_features, hidden_features)
self.w2 = nn.Linear(hidden_features, in_features)
def forward(self, x):
x = self.norm(x)
x = self.act(self.w0(x)) * self.w1(x)
x = self.w2(x)
return x