This commit is contained in:
2025-08-20 17:05:36 +08:00
parent 6ae2fb5868
commit 82140eff2c
259 changed files with 505375 additions and 251 deletions

View File

@@ -0,0 +1,13 @@
# from f5_tts.model.cfm import CFM
#
# from f5_tts.model.backbones.unett import UNetT
from GPT_SoVITS.f5_tts.model.backbones.dit import DiT
# from f5_tts.model.backbones.dit import DiTNoCond
# from f5_tts.model.backbones.dit import DiTNoCondNoT
# from f5_tts.model.backbones.mmdit import MMDiT
# from f5_tts.model.trainer import Trainer
# __all__ = ["CFM", "UNetT", "DiT", "MMDiT", "Trainer"]
# __all__ = ["CFM", "UNetT", "DiTNoCond","DiT", "MMDiT"]

View File

@@ -0,0 +1,20 @@
## Backbones quick introduction
### unett.py
- flat unet transformer
- structure same as in e2-tts & voicebox paper except using rotary pos emb
- update: allow possible abs pos emb & convnextv2 blocks for embedded text before concat
### dit.py
- adaln-zero dit
- embedded timestep as condition
- concatted noised_input + masked_cond + embedded_text, linear proj in
- possible abs pos emb & convnextv2 blocks for embedded text before concat
- possible long skip connection (first layer to last layer)
### mmdit.py
- sd3 structure
- timestep as condition
- left stream: text embedded and applied a abs pos emb
- right stream: masked_cond & noised_input concatted and with same conv pos emb as unett

View File

@@ -0,0 +1,194 @@
"""
ein notation:
b - batch
n - sequence
nt - text sequence
nw - raw wave length
d - dimension
"""
from __future__ import annotations
import torch
from torch import nn
from torch.utils.checkpoint import checkpoint
from x_transformers.x_transformers import RotaryEmbedding
from GPT_SoVITS.f5_tts.model.modules import (
TimestepEmbedding,
ConvNeXtV2Block,
ConvPositionEmbedding,
DiTBlock,
AdaLayerNormZero_Final,
precompute_freqs_cis,
get_pos_embed_indices,
)
from module.commons import sequence_mask
class TextEmbedding(nn.Module):
def __init__(self, text_dim, conv_layers=0, conv_mult=2):
super().__init__()
if conv_layers > 0:
self.extra_modeling = True
self.precompute_max_pos = 4096 # ~44s of 24khz audio
self.register_buffer("freqs_cis", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)
self.text_blocks = nn.Sequential(
*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]
)
else:
self.extra_modeling = False
def forward(self, text: int["b nt"], seq_len, drop_text=False): # noqa: F722
batch, text_len = text.shape[0], text.shape[1]
if drop_text: # cfg for text
text = torch.zeros_like(text)
# possible extra modeling
if self.extra_modeling:
# sinus pos emb
batch_start = torch.zeros((batch,), dtype=torch.long)
pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)
text_pos_embed = self.freqs_cis[pos_idx]
# print(23333333,text.shape,text_pos_embed.shape)#torch.Size([7, 465, 256]) torch.Size([7, 465, 256])
text = text + text_pos_embed
# convnextv2 blocks
text = self.text_blocks(text)
return text
# noised input audio and context mixing embedding
class InputEmbedding(nn.Module):
def __init__(self, mel_dim, text_dim, out_dim):
super().__init__()
self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)
self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)
def forward(self, x: float["b n d"], cond: float["b n d"], text_embed: float["b n d"], drop_audio_cond=False): # noqa: F722
if drop_audio_cond: # cfg for cond audio
cond = torch.zeros_like(cond)
x = self.proj(torch.cat((x, cond, text_embed), dim=-1))
x = self.conv_pos_embed(x) + x
return x
# Transformer backbone using DiT blocks
class DiT(nn.Module):
def __init__(
self,
*,
dim,
depth=8,
heads=8,
dim_head=64,
dropout=0.1,
ff_mult=4,
mel_dim=100,
text_dim=None,
conv_layers=0,
long_skip_connection=False,
):
super().__init__()
self.time_embed = TimestepEmbedding(dim)
self.d_embed = TimestepEmbedding(dim)
if text_dim is None:
text_dim = mel_dim
self.text_embed = TextEmbedding(text_dim, conv_layers=conv_layers)
self.input_embed = InputEmbedding(mel_dim, text_dim, dim)
self.rotary_embed = RotaryEmbedding(dim_head)
self.dim = dim
self.depth = depth
self.transformer_blocks = nn.ModuleList(
[DiTBlock(dim=dim, heads=heads, dim_head=dim_head, ff_mult=ff_mult, dropout=dropout) for _ in range(depth)]
)
self.long_skip_connection = nn.Linear(dim * 2, dim, bias=False) if long_skip_connection else None
self.norm_out = AdaLayerNormZero_Final(dim) # final modulation
self.proj_out = nn.Linear(dim, mel_dim)
def ckpt_wrapper(self, module):
# https://github.com/chuanyangjin/fast-DiT/blob/main/models.py
def ckpt_forward(*inputs):
outputs = module(*inputs)
return outputs
return ckpt_forward
def forward( # x, prompt_x, x_lens, t, style,cond
self, # d is channel,n is T
x0: float["b n d"], # nosied input audio # noqa: F722
cond0: float["b n d"], # masked cond audio # noqa: F722
x_lens,
time: float["b"] | float[""], # time step # noqa: F821 F722
dt_base_bootstrap,
text0, # : int["b nt"] # noqa: F722#####condition feature
use_grad_ckpt=False, # bool
###no-use
drop_audio_cond=False, # cfg for cond audio
drop_text=False, # cfg for text
# mask: bool["b n"] | None = None, # noqa: F722
infer=False, # bool
text_cache=None, # torch tensor as text_embed
dt_cache=None, # torch tensor as dt
):
x = x0.transpose(2, 1)
cond = cond0.transpose(2, 1)
text = text0.transpose(2, 1)
mask = sequence_mask(x_lens, max_length=x.size(1)).to(x.device)
batch, seq_len = x.shape[0], x.shape[1]
if time.ndim == 0:
time = time.repeat(batch)
# t: conditioning time, c: context (text + masked cond audio), x: noised input audio
t = self.time_embed(time)
if infer and dt_cache is not None:
dt = dt_cache
else:
dt = self.d_embed(dt_base_bootstrap)
t += dt
if infer and text_cache is not None:
text_embed = text_cache
else:
text_embed = self.text_embed(text, seq_len, drop_text=drop_text) ###need to change
x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)
rope = self.rotary_embed.forward_from_seq_len(seq_len)
if self.long_skip_connection is not None:
residual = x
for block in self.transformer_blocks:
if use_grad_ckpt:
x = checkpoint(self.ckpt_wrapper(block), x, t, mask, rope, use_reentrant=False)
else:
x = block(x, t, mask=mask, rope=rope)
if self.long_skip_connection is not None:
x = self.long_skip_connection(torch.cat((x, residual), dim=-1))
x = self.norm_out(x, t)
output = self.proj_out(x)
if infer:
return output, text_embed, dt
else:
return output

View File

@@ -0,0 +1,146 @@
"""
ein notation:
b - batch
n - sequence
nt - text sequence
nw - raw wave length
d - dimension
"""
from __future__ import annotations
import torch
from torch import nn
from x_transformers.x_transformers import RotaryEmbedding
from f5_tts.model.modules import (
TimestepEmbedding,
ConvPositionEmbedding,
MMDiTBlock,
AdaLayerNormZero_Final,
precompute_freqs_cis,
get_pos_embed_indices,
)
# text embedding
class TextEmbedding(nn.Module):
def __init__(self, out_dim, text_num_embeds):
super().__init__()
self.text_embed = nn.Embedding(text_num_embeds + 1, out_dim) # will use 0 as filler token
self.precompute_max_pos = 1024
self.register_buffer("freqs_cis", precompute_freqs_cis(out_dim, self.precompute_max_pos), persistent=False)
def forward(self, text: int["b nt"], drop_text=False) -> int["b nt d"]: # noqa: F722
text = text + 1
if drop_text:
text = torch.zeros_like(text)
text = self.text_embed(text)
# sinus pos emb
batch_start = torch.zeros((text.shape[0],), dtype=torch.long)
batch_text_len = text.shape[1]
pos_idx = get_pos_embed_indices(batch_start, batch_text_len, max_pos=self.precompute_max_pos)
text_pos_embed = self.freqs_cis[pos_idx]
text = text + text_pos_embed
return text
# noised input & masked cond audio embedding
class AudioEmbedding(nn.Module):
def __init__(self, in_dim, out_dim):
super().__init__()
self.linear = nn.Linear(2 * in_dim, out_dim)
self.conv_pos_embed = ConvPositionEmbedding(out_dim)
def forward(self, x: float["b n d"], cond: float["b n d"], drop_audio_cond=False): # noqa: F722
if drop_audio_cond:
cond = torch.zeros_like(cond)
x = torch.cat((x, cond), dim=-1)
x = self.linear(x)
x = self.conv_pos_embed(x) + x
return x
# Transformer backbone using MM-DiT blocks
class MMDiT(nn.Module):
def __init__(
self,
*,
dim,
depth=8,
heads=8,
dim_head=64,
dropout=0.1,
ff_mult=4,
text_num_embeds=256,
mel_dim=100,
):
super().__init__()
self.time_embed = TimestepEmbedding(dim)
self.text_embed = TextEmbedding(dim, text_num_embeds)
self.audio_embed = AudioEmbedding(mel_dim, dim)
self.rotary_embed = RotaryEmbedding(dim_head)
self.dim = dim
self.depth = depth
self.transformer_blocks = nn.ModuleList(
[
MMDiTBlock(
dim=dim,
heads=heads,
dim_head=dim_head,
dropout=dropout,
ff_mult=ff_mult,
context_pre_only=i == depth - 1,
)
for i in range(depth)
]
)
self.norm_out = AdaLayerNormZero_Final(dim) # final modulation
self.proj_out = nn.Linear(dim, mel_dim)
def forward(
self,
x: float["b n d"], # nosied input audio # noqa: F722
cond: float["b n d"], # masked cond audio # noqa: F722
text: int["b nt"], # text # noqa: F722
time: float["b"] | float[""], # time step # noqa: F821 F722
drop_audio_cond, # cfg for cond audio
drop_text, # cfg for text
mask: bool["b n"] | None = None, # noqa: F722
):
batch = x.shape[0]
if time.ndim == 0:
time = time.repeat(batch)
# t: conditioning (time), c: context (text + masked cond audio), x: noised input audio
t = self.time_embed(time)
c = self.text_embed(text, drop_text=drop_text)
x = self.audio_embed(x, cond, drop_audio_cond=drop_audio_cond)
seq_len = x.shape[1]
text_len = text.shape[1]
rope_audio = self.rotary_embed.forward_from_seq_len(seq_len)
rope_text = self.rotary_embed.forward_from_seq_len(text_len)
for block in self.transformer_blocks:
c, x = block(x, c, t, mask=mask, rope=rope_audio, c_rope=rope_text)
x = self.norm_out(x, t)
output = self.proj_out(x)
return output

View File

@@ -0,0 +1,219 @@
"""
ein notation:
b - batch
n - sequence
nt - text sequence
nw - raw wave length
d - dimension
"""
from __future__ import annotations
from typing import Literal
import torch
from torch import nn
import torch.nn.functional as F
from x_transformers import RMSNorm
from x_transformers.x_transformers import RotaryEmbedding
from f5_tts.model.modules import (
TimestepEmbedding,
ConvNeXtV2Block,
ConvPositionEmbedding,
Attention,
AttnProcessor,
FeedForward,
precompute_freqs_cis,
get_pos_embed_indices,
)
# Text embedding
class TextEmbedding(nn.Module):
def __init__(self, text_num_embeds, text_dim, conv_layers=0, conv_mult=2):
super().__init__()
self.text_embed = nn.Embedding(text_num_embeds + 1, text_dim) # use 0 as filler token
if conv_layers > 0:
self.extra_modeling = True
self.precompute_max_pos = 4096 # ~44s of 24khz audio
self.register_buffer("freqs_cis", precompute_freqs_cis(text_dim, self.precompute_max_pos), persistent=False)
self.text_blocks = nn.Sequential(
*[ConvNeXtV2Block(text_dim, text_dim * conv_mult) for _ in range(conv_layers)]
)
else:
self.extra_modeling = False
def forward(self, text: int["b nt"], seq_len, drop_text=False): # noqa: F722
text = text + 1 # use 0 as filler token. preprocess of batch pad -1, see list_str_to_idx()
text = text[:, :seq_len] # curtail if character tokens are more than the mel spec tokens
batch, text_len = text.shape[0], text.shape[1]
text = F.pad(text, (0, seq_len - text_len), value=0)
if drop_text: # cfg for text
text = torch.zeros_like(text)
text = self.text_embed(text) # b n -> b n d
# possible extra modeling
if self.extra_modeling:
# sinus pos emb
batch_start = torch.zeros((batch,), dtype=torch.long)
pos_idx = get_pos_embed_indices(batch_start, seq_len, max_pos=self.precompute_max_pos)
text_pos_embed = self.freqs_cis[pos_idx]
text = text + text_pos_embed
# convnextv2 blocks
text = self.text_blocks(text)
return text
# noised input audio and context mixing embedding
class InputEmbedding(nn.Module):
def __init__(self, mel_dim, text_dim, out_dim):
super().__init__()
self.proj = nn.Linear(mel_dim * 2 + text_dim, out_dim)
self.conv_pos_embed = ConvPositionEmbedding(dim=out_dim)
def forward(self, x: float["b n d"], cond: float["b n d"], text_embed: float["b n d"], drop_audio_cond=False): # noqa: F722
if drop_audio_cond: # cfg for cond audio
cond = torch.zeros_like(cond)
x = self.proj(torch.cat((x, cond, text_embed), dim=-1))
x = self.conv_pos_embed(x) + x
return x
# Flat UNet Transformer backbone
class UNetT(nn.Module):
def __init__(
self,
*,
dim,
depth=8,
heads=8,
dim_head=64,
dropout=0.1,
ff_mult=4,
mel_dim=100,
text_num_embeds=256,
text_dim=None,
conv_layers=0,
skip_connect_type: Literal["add", "concat", "none"] = "concat",
):
super().__init__()
assert depth % 2 == 0, "UNet-Transformer's depth should be even."
self.time_embed = TimestepEmbedding(dim)
if text_dim is None:
text_dim = mel_dim
self.text_embed = TextEmbedding(text_num_embeds, text_dim, conv_layers=conv_layers)
self.input_embed = InputEmbedding(mel_dim, text_dim, dim)
self.rotary_embed = RotaryEmbedding(dim_head)
# transformer layers & skip connections
self.dim = dim
self.skip_connect_type = skip_connect_type
needs_skip_proj = skip_connect_type == "concat"
self.depth = depth
self.layers = nn.ModuleList([])
for idx in range(depth):
is_later_half = idx >= (depth // 2)
attn_norm = RMSNorm(dim)
attn = Attention(
processor=AttnProcessor(),
dim=dim,
heads=heads,
dim_head=dim_head,
dropout=dropout,
)
ff_norm = RMSNorm(dim)
ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate="tanh")
skip_proj = nn.Linear(dim * 2, dim, bias=False) if needs_skip_proj and is_later_half else None
self.layers.append(
nn.ModuleList(
[
skip_proj,
attn_norm,
attn,
ff_norm,
ff,
]
)
)
self.norm_out = RMSNorm(dim)
self.proj_out = nn.Linear(dim, mel_dim)
def forward(
self,
x: float["b n d"], # nosied input audio # noqa: F722
cond: float["b n d"], # masked cond audio # noqa: F722
text: int["b nt"], # text # noqa: F722
time: float["b"] | float[""], # time step # noqa: F821 F722
drop_audio_cond, # cfg for cond audio
drop_text, # cfg for text
mask: bool["b n"] | None = None, # noqa: F722
):
batch, seq_len = x.shape[0], x.shape[1]
if time.ndim == 0:
time = time.repeat(batch)
# t: conditioning time, c: context (text + masked cond audio), x: noised input audio
t = self.time_embed(time)
text_embed = self.text_embed(text, seq_len, drop_text=drop_text)
x = self.input_embed(x, cond, text_embed, drop_audio_cond=drop_audio_cond)
# postfix time t to input x, [b n d] -> [b n+1 d]
x = torch.cat([t.unsqueeze(1), x], dim=1) # pack t to x
if mask is not None:
mask = F.pad(mask, (1, 0), value=1)
rope = self.rotary_embed.forward_from_seq_len(seq_len + 1)
# flat unet transformer
skip_connect_type = self.skip_connect_type
skips = []
for idx, (maybe_skip_proj, attn_norm, attn, ff_norm, ff) in enumerate(self.layers):
layer = idx + 1
# skip connection logic
is_first_half = layer <= (self.depth // 2)
is_later_half = not is_first_half
if is_first_half:
skips.append(x)
if is_later_half:
skip = skips.pop()
if skip_connect_type == "concat":
x = torch.cat((x, skip), dim=-1)
x = maybe_skip_proj(x)
elif skip_connect_type == "add":
x = x + skip
# attention and feedforward blocks
x = attn(attn_norm(x), rope=rope, mask=mask) + x
x = ff(ff_norm(x)) + x
assert len(skips) == 0
x = self.norm_out(x)[:, 1:, :] # unpack t from x
return self.proj_out(x)

View File

@@ -0,0 +1,666 @@
"""
ein notation:
b - batch
n - sequence
nt - text sequence
nw - raw wave length
d - dimension
"""
from __future__ import annotations
import math
from typing import Optional
import torch
import torch.nn.functional as F
import torchaudio
from librosa.filters import mel as librosa_mel_fn
from torch import nn
from x_transformers.x_transformers import apply_rotary_pos_emb
# raw wav to mel spec
mel_basis_cache = {}
hann_window_cache = {}
def get_bigvgan_mel_spectrogram(
waveform,
n_fft=1024,
n_mel_channels=100,
target_sample_rate=24000,
hop_length=256,
win_length=1024,
fmin=0,
fmax=None,
center=False,
): # Copy from https://github.com/NVIDIA/BigVGAN/tree/main
device = waveform.device
key = f"{n_fft}_{n_mel_channels}_{target_sample_rate}_{hop_length}_{win_length}_{fmin}_{fmax}_{device}"
if key not in mel_basis_cache:
mel = librosa_mel_fn(sr=target_sample_rate, n_fft=n_fft, n_mels=n_mel_channels, fmin=fmin, fmax=fmax)
mel_basis_cache[key] = torch.from_numpy(mel).float().to(device) # TODO: why they need .float()?
hann_window_cache[key] = torch.hann_window(win_length).to(device)
mel_basis = mel_basis_cache[key]
hann_window = hann_window_cache[key]
padding = (n_fft - hop_length) // 2
waveform = torch.nn.functional.pad(waveform.unsqueeze(1), (padding, padding), mode="reflect").squeeze(1)
spec = torch.stft(
waveform,
n_fft,
hop_length=hop_length,
win_length=win_length,
window=hann_window,
center=center,
pad_mode="reflect",
normalized=False,
onesided=True,
return_complex=True,
)
spec = torch.sqrt(torch.view_as_real(spec).pow(2).sum(-1) + 1e-9)
mel_spec = torch.matmul(mel_basis, spec)
mel_spec = torch.log(torch.clamp(mel_spec, min=1e-5))
return mel_spec
def get_vocos_mel_spectrogram(
waveform,
n_fft=1024,
n_mel_channels=100,
target_sample_rate=24000,
hop_length=256,
win_length=1024,
):
mel_stft = torchaudio.transforms.MelSpectrogram(
sample_rate=target_sample_rate,
n_fft=n_fft,
win_length=win_length,
hop_length=hop_length,
n_mels=n_mel_channels,
power=1,
center=True,
normalized=False,
norm=None,
).to(waveform.device)
if len(waveform.shape) == 3:
waveform = waveform.squeeze(1) # 'b 1 nw -> b nw'
assert len(waveform.shape) == 2
mel = mel_stft(waveform)
mel = mel.clamp(min=1e-5).log()
return mel
class MelSpec(nn.Module):
def __init__(
self,
n_fft=1024,
hop_length=256,
win_length=1024,
n_mel_channels=100,
target_sample_rate=24_000,
mel_spec_type="vocos",
):
super().__init__()
assert mel_spec_type in ["vocos", "bigvgan"], print("We only support two extract mel backend: vocos or bigvgan")
self.n_fft = n_fft
self.hop_length = hop_length
self.win_length = win_length
self.n_mel_channels = n_mel_channels
self.target_sample_rate = target_sample_rate
if mel_spec_type == "vocos":
self.extractor = get_vocos_mel_spectrogram
elif mel_spec_type == "bigvgan":
self.extractor = get_bigvgan_mel_spectrogram
self.register_buffer("dummy", torch.tensor(0), persistent=False)
def forward(self, wav):
if self.dummy.device != wav.device:
self.to(wav.device)
mel = self.extractor(
waveform=wav,
n_fft=self.n_fft,
n_mel_channels=self.n_mel_channels,
target_sample_rate=self.target_sample_rate,
hop_length=self.hop_length,
win_length=self.win_length,
)
return mel
# sinusoidal position embedding
class SinusPositionEmbedding(nn.Module):
def __init__(self, dim):
super().__init__()
self.dim = dim
def forward(self, x, scale=1000):
device = x.device
half_dim = self.dim // 2
emb = math.log(10000) / (half_dim - 1)
emb = torch.exp(torch.arange(half_dim, device=device).float() * -emb)
emb = scale * x.unsqueeze(1) * emb.unsqueeze(0)
emb = torch.cat((emb.sin(), emb.cos()), dim=-1)
return emb
# convolutional position embedding
class ConvPositionEmbedding(nn.Module):
def __init__(self, dim, kernel_size=31, groups=16):
super().__init__()
assert kernel_size % 2 != 0
self.conv1d = nn.Sequential(
nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),
nn.Mish(),
nn.Conv1d(dim, dim, kernel_size, groups=groups, padding=kernel_size // 2),
nn.Mish(),
)
def forward(self, x: float["b n d"], mask: bool["b n"] | None = None): # noqa: F722
if mask is not None:
mask = mask[..., None]
x = x.masked_fill(~mask, 0.0)
x = x.permute(0, 2, 1)
x = self.conv1d(x)
out = x.permute(0, 2, 1)
if mask is not None:
out = out.masked_fill(~mask, 0.0)
return out
# rotary positional embedding related
def precompute_freqs_cis(dim: int, end: int, theta: float = 10000.0, theta_rescale_factor=1.0):
# proposed by reddit user bloc97, to rescale rotary embeddings to longer sequence length without fine-tuning
# has some connection to NTK literature
# https://www.reddit.com/r/LocalLLaMA/comments/14lz7j5/ntkaware_scaled_rope_allows_llama_models_to_have/
# https://github.com/lucidrains/rotary-embedding-torch/blob/main/rotary_embedding_torch/rotary_embedding_torch.py
theta *= theta_rescale_factor ** (dim / (dim - 2))
freqs = 1.0 / (theta ** (torch.arange(0, dim, 2)[: (dim // 2)].float() / dim))
t = torch.arange(end, device=freqs.device) # type: ignore
freqs = torch.outer(t, freqs).float() # type: ignore
freqs_cos = torch.cos(freqs) # real part
freqs_sin = torch.sin(freqs) # imaginary part
return torch.cat([freqs_cos, freqs_sin], dim=-1)
def get_pos_embed_indices(start, length, max_pos, scale=1.0):
# length = length if isinstance(length, int) else length.max()
scale = scale * torch.ones_like(start, dtype=torch.float32) # in case scale is a scalar
pos = (
start.unsqueeze(1)
+ (torch.arange(length, device=start.device, dtype=torch.float32).unsqueeze(0) * scale.unsqueeze(1)).long()
)
# avoid extra long error.
pos = torch.where(pos < max_pos, pos, max_pos - 1)
return pos
# Global Response Normalization layer (Instance Normalization ?)
class GRN(nn.Module):
def __init__(self, dim):
super().__init__()
self.gamma = nn.Parameter(torch.zeros(1, 1, dim))
self.beta = nn.Parameter(torch.zeros(1, 1, dim))
def forward(self, x):
Gx = torch.norm(x, p=2, dim=1, keepdim=True)
Nx = Gx / (Gx.mean(dim=-1, keepdim=True) + 1e-6)
return self.gamma * (x * Nx) + self.beta + x
# ConvNeXt-V2 Block https://github.com/facebookresearch/ConvNeXt-V2/blob/main/models/convnextv2.py
# ref: https://github.com/bfs18/e2_tts/blob/main/rfwave/modules.py#L108
class ConvNeXtV2Block(nn.Module):
def __init__(
self,
dim: int,
intermediate_dim: int,
dilation: int = 1,
):
super().__init__()
padding = (dilation * (7 - 1)) // 2
self.dwconv = nn.Conv1d(
dim, dim, kernel_size=7, padding=padding, groups=dim, dilation=dilation
) # depthwise conv
self.norm = nn.LayerNorm(dim, eps=1e-6)
self.pwconv1 = nn.Linear(dim, intermediate_dim) # pointwise/1x1 convs, implemented with linear layers
self.act = nn.GELU()
self.grn = GRN(intermediate_dim)
self.pwconv2 = nn.Linear(intermediate_dim, dim)
def forward(self, x: torch.Tensor) -> torch.Tensor:
residual = x
x = x.transpose(1, 2) # b n d -> b d n
x = self.dwconv(x)
x = x.transpose(1, 2) # b d n -> b n d
x = self.norm(x)
x = self.pwconv1(x)
x = self.act(x)
x = self.grn(x)
x = self.pwconv2(x)
return residual + x
# AdaLayerNormZero
# return with modulated x for attn input, and params for later mlp modulation
class AdaLayerNormZero(nn.Module):
def __init__(self, dim):
super().__init__()
self.silu = nn.SiLU()
self.linear = nn.Linear(dim, dim * 6)
self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
def forward(self, x, emb=None):
emb = self.linear(self.silu(emb))
shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = torch.chunk(emb, 6, dim=1)
x = self.norm(x) * (1 + scale_msa[:, None]) + shift_msa[:, None]
return x, gate_msa, shift_mlp, scale_mlp, gate_mlp
# AdaLayerNormZero for final layer
# return only with modulated x for attn input, cuz no more mlp modulation
class AdaLayerNormZero_Final(nn.Module):
def __init__(self, dim):
super().__init__()
self.silu = nn.SiLU()
self.linear = nn.Linear(dim, dim * 2)
self.norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
def forward(self, x, emb):
emb = self.linear(self.silu(emb))
scale, shift = torch.chunk(emb, 2, dim=1)
x = self.norm(x) * (1 + scale)[:, None, :] + shift[:, None, :]
return x
# FeedForward
class FeedForward(nn.Module):
def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, approximate: str = "none"):
super().__init__()
inner_dim = int(dim * mult)
dim_out = dim_out if dim_out is not None else dim
activation = nn.GELU(approximate=approximate)
project_in = nn.Sequential(nn.Linear(dim, inner_dim), activation)
self.ff = nn.Sequential(project_in, nn.Dropout(dropout), nn.Linear(inner_dim, dim_out))
def forward(self, x):
return self.ff(x)
# Attention with possible joint part
# modified from diffusers/src/diffusers/models/attention_processor.py
class Attention(nn.Module):
def __init__(
self,
processor: JointAttnProcessor | AttnProcessor,
dim: int,
heads: int = 8,
dim_head: int = 64,
dropout: float = 0.0,
context_dim: Optional[int] = None, # if not None -> joint attention
context_pre_only=None,
):
super().__init__()
if not hasattr(F, "scaled_dot_product_attention"):
raise ImportError("Attention equires PyTorch 2.0, to use it, please upgrade PyTorch to 2.0.")
self.processor = processor
self.dim = dim
self.heads = heads
self.inner_dim = dim_head * heads
self.dropout = dropout
self.context_dim = context_dim
self.context_pre_only = context_pre_only
self.to_q = nn.Linear(dim, self.inner_dim)
self.to_k = nn.Linear(dim, self.inner_dim)
self.to_v = nn.Linear(dim, self.inner_dim)
if self.context_dim is not None:
self.to_k_c = nn.Linear(context_dim, self.inner_dim)
self.to_v_c = nn.Linear(context_dim, self.inner_dim)
if self.context_pre_only is not None:
self.to_q_c = nn.Linear(context_dim, self.inner_dim)
self.to_out = nn.ModuleList([])
self.to_out.append(nn.Linear(self.inner_dim, dim))
self.to_out.append(nn.Dropout(dropout))
if self.context_pre_only is not None and not self.context_pre_only:
self.to_out_c = nn.Linear(self.inner_dim, dim)
def forward(
self,
x: float["b n d"], # noised input x # noqa: F722
c: float["b n d"] = None, # context c # noqa: F722
mask: bool["b n"] | None = None, # noqa: F722
rope=None, # rotary position embedding for x
c_rope=None, # rotary position embedding for c
) -> torch.Tensor:
if c is not None:
return self.processor(self, x, c=c, mask=mask, rope=rope, c_rope=c_rope)
else:
return self.processor(self, x, mask=mask, rope=rope)
# Attention processor
# from torch.nn.attention import SDPBackend
# torch.backends.cuda.enable_flash_sdp(True)
class AttnProcessor:
def __init__(self):
pass
def __call__(
self,
attn: Attention,
x: float["b n d"], # noised input x # noqa: F722
mask: bool["b n"] | None = None, # noqa: F722
rope=None, # rotary position embedding
) -> torch.FloatTensor:
batch_size = x.shape[0]
# `sample` projections.
query = attn.to_q(x)
key = attn.to_k(x)
value = attn.to_v(x)
# apply rotary position embedding
if rope is not None:
freqs, xpos_scale = rope
q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)
query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)
key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)
# attention
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)
# mask. e.g. inference got a batch with different target durations, mask out the padding
if mask is not None:
attn_mask = mask
attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'
# print(3433333333,attn_mask.shape)
attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])
else:
attn_mask = None
# with torch.nn.attention.sdpa_kernel(backends=[SDPBackend.EFFICIENT_ATTENTION]):
# with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=False, enable_mem_efficient=True):
# with torch.backends.cuda.sdp_kernel(enable_flash=True, enable_math=True, enable_mem_efficient=False):
# print(torch.backends.cuda.flash_sdp_enabled())
# print(torch.backends.cuda.mem_efficient_sdp_enabled())
# print(torch.backends.cuda.math_sdp_enabled())
x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)
x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
x = x.to(query.dtype)
# linear proj
x = attn.to_out[0](x)
# dropout
x = attn.to_out[1](x)
if mask is not None:
mask = mask.unsqueeze(-1)
x = x.masked_fill(~mask, 0.0)
return x
# Joint Attention processor for MM-DiT
# modified from diffusers/src/diffusers/models/attention_processor.py
class JointAttnProcessor:
def __init__(self):
pass
def __call__(
self,
attn: Attention,
x: float["b n d"], # noised input x # noqa: F722
c: float["b nt d"] = None, # context c, here text # noqa: F722
mask: bool["b n"] | None = None, # noqa: F722
rope=None, # rotary position embedding for x
c_rope=None, # rotary position embedding for c
) -> torch.FloatTensor:
residual = x
batch_size = c.shape[0]
# `sample` projections.
query = attn.to_q(x)
key = attn.to_k(x)
value = attn.to_v(x)
# `context` projections.
c_query = attn.to_q_c(c)
c_key = attn.to_k_c(c)
c_value = attn.to_v_c(c)
# apply rope for context and noised input independently
if rope is not None:
freqs, xpos_scale = rope
q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)
query = apply_rotary_pos_emb(query, freqs, q_xpos_scale)
key = apply_rotary_pos_emb(key, freqs, k_xpos_scale)
if c_rope is not None:
freqs, xpos_scale = c_rope
q_xpos_scale, k_xpos_scale = (xpos_scale, xpos_scale**-1.0) if xpos_scale is not None else (1.0, 1.0)
c_query = apply_rotary_pos_emb(c_query, freqs, q_xpos_scale)
c_key = apply_rotary_pos_emb(c_key, freqs, k_xpos_scale)
# attention
query = torch.cat([query, c_query], dim=1)
key = torch.cat([key, c_key], dim=1)
value = torch.cat([value, c_value], dim=1)
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)
# mask. e.g. inference got a batch with different target durations, mask out the padding
if mask is not None:
attn_mask = F.pad(mask, (0, c.shape[1]), value=True) # no mask for c (text)
attn_mask = attn_mask.unsqueeze(1).unsqueeze(1) # 'b n -> b 1 1 n'
attn_mask = attn_mask.expand(batch_size, attn.heads, query.shape[-2], key.shape[-2])
else:
attn_mask = None
x = F.scaled_dot_product_attention(query, key, value, attn_mask=attn_mask, dropout_p=0.0, is_causal=False)
x = x.transpose(1, 2).reshape(batch_size, -1, attn.heads * head_dim)
x = x.to(query.dtype)
# Split the attention outputs.
x, c = (
x[:, : residual.shape[1]],
x[:, residual.shape[1] :],
)
# linear proj
x = attn.to_out[0](x)
# dropout
x = attn.to_out[1](x)
if not attn.context_pre_only:
c = attn.to_out_c(c)
if mask is not None:
mask = mask.unsqueeze(-1)
x = x.masked_fill(~mask, 0.0)
# c = c.masked_fill(~mask, 0.) # no mask for c (text)
return x, c
# DiT Block
class DiTBlock(nn.Module):
def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1):
super().__init__()
self.attn_norm = AdaLayerNormZero(dim)
self.attn = Attention(
processor=AttnProcessor(),
dim=dim,
heads=heads,
dim_head=dim_head,
dropout=dropout,
)
self.ff_norm = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
self.ff = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate="tanh")
def forward(self, x, t, mask=None, rope=None): # x: noised input, t: time embedding
# pre-norm & modulation for attention input
norm, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.attn_norm(x, emb=t)
# attention
attn_output = self.attn(x=norm, mask=mask, rope=rope)
# process attention output for input x
x = x + gate_msa.unsqueeze(1) * attn_output
norm = self.ff_norm(x) * (1 + scale_mlp[:, None]) + shift_mlp[:, None]
ff_output = self.ff(norm)
x = x + gate_mlp.unsqueeze(1) * ff_output
return x
# MMDiT Block https://arxiv.org/abs/2403.03206
class MMDiTBlock(nn.Module):
r"""
modified from diffusers/src/diffusers/models/attention.py
notes.
_c: context related. text, cond, etc. (left part in sd3 fig2.b)
_x: noised input related. (right part)
context_pre_only: last layer only do prenorm + modulation cuz no more ffn
"""
def __init__(self, dim, heads, dim_head, ff_mult=4, dropout=0.1, context_pre_only=False):
super().__init__()
self.context_pre_only = context_pre_only
self.attn_norm_c = AdaLayerNormZero_Final(dim) if context_pre_only else AdaLayerNormZero(dim)
self.attn_norm_x = AdaLayerNormZero(dim)
self.attn = Attention(
processor=JointAttnProcessor(),
dim=dim,
heads=heads,
dim_head=dim_head,
dropout=dropout,
context_dim=dim,
context_pre_only=context_pre_only,
)
if not context_pre_only:
self.ff_norm_c = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
self.ff_c = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate="tanh")
else:
self.ff_norm_c = None
self.ff_c = None
self.ff_norm_x = nn.LayerNorm(dim, elementwise_affine=False, eps=1e-6)
self.ff_x = FeedForward(dim=dim, mult=ff_mult, dropout=dropout, approximate="tanh")
def forward(self, x, c, t, mask=None, rope=None, c_rope=None): # x: noised input, c: context, t: time embedding
# pre-norm & modulation for attention input
if self.context_pre_only:
norm_c = self.attn_norm_c(c, t)
else:
norm_c, c_gate_msa, c_shift_mlp, c_scale_mlp, c_gate_mlp = self.attn_norm_c(c, emb=t)
norm_x, x_gate_msa, x_shift_mlp, x_scale_mlp, x_gate_mlp = self.attn_norm_x(x, emb=t)
# attention
x_attn_output, c_attn_output = self.attn(x=norm_x, c=norm_c, mask=mask, rope=rope, c_rope=c_rope)
# process attention output for context c
if self.context_pre_only:
c = None
else: # if not last layer
c = c + c_gate_msa.unsqueeze(1) * c_attn_output
norm_c = self.ff_norm_c(c) * (1 + c_scale_mlp[:, None]) + c_shift_mlp[:, None]
c_ff_output = self.ff_c(norm_c)
c = c + c_gate_mlp.unsqueeze(1) * c_ff_output
# process attention output for input x
x = x + x_gate_msa.unsqueeze(1) * x_attn_output
norm_x = self.ff_norm_x(x) * (1 + x_scale_mlp[:, None]) + x_shift_mlp[:, None]
x_ff_output = self.ff_x(norm_x)
x = x + x_gate_mlp.unsqueeze(1) * x_ff_output
return c, x
# time step conditioning embedding
class TimestepEmbedding(nn.Module):
def __init__(self, dim, freq_embed_dim=256):
super().__init__()
self.time_embed = SinusPositionEmbedding(freq_embed_dim)
self.time_mlp = nn.Sequential(nn.Linear(freq_embed_dim, dim), nn.SiLU(), nn.Linear(dim, dim))
def forward(self, timestep: float["b"]): # noqa: F821
time_hidden = self.time_embed(timestep)
time_hidden = time_hidden.to(timestep.dtype)
time = self.time_mlp(time_hidden) # b d
return time