DL-Art-School/codes/models/audio/music/flat_diffusion.py

333 lines
16 KiB
Python
Raw Normal View History

2022-05-20 17:01:48 +00:00
import random
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch import autocast
2022-05-20 23:04:22 +00:00
from models.arch_util import ResBlock
2022-05-20 17:01:48 +00:00
from models.diffusion.nn import timestep_embedding, normalization, zero_module, conv_nd, linear
from models.diffusion.unet_diffusion import AttentionBlock, TimestepEmbedSequential, TimestepBlock
from trainer.networks import register_model
from utils.util import checkpoint
def is_latent(t):
return t.dtype == torch.float
def is_sequence(t):
return t.dtype == torch.long
2022-05-20 23:04:22 +00:00
class TimestepResBlock(TimestepBlock):
2022-05-20 17:01:48 +00:00
def __init__(
self,
channels,
emb_channels,
dropout,
out_channels=None,
dims=2,
kernel_size=3,
efficient_config=True,
use_scale_shift_norm=False,
):
super().__init__()
self.channels = channels
self.emb_channels = emb_channels
self.dropout = dropout
self.out_channels = out_channels or channels
self.use_scale_shift_norm = use_scale_shift_norm
padding = {1: 0, 3: 1, 5: 2}[kernel_size]
eff_kernel = 1 if efficient_config else 3
eff_padding = 0 if efficient_config else 1
self.in_layers = nn.Sequential(
normalization(channels),
nn.SiLU(),
conv_nd(dims, channels, self.out_channels, eff_kernel, padding=eff_padding),
)
self.emb_layers = nn.Sequential(
nn.SiLU(),
linear(
emb_channels,
2 * self.out_channels if use_scale_shift_norm else self.out_channels,
),
)
self.out_layers = nn.Sequential(
normalization(self.out_channels),
nn.SiLU(),
nn.Dropout(p=dropout),
zero_module(
conv_nd(dims, self.out_channels, self.out_channels, kernel_size, padding=padding)
),
)
if self.out_channels == channels:
self.skip_connection = nn.Identity()
else:
self.skip_connection = conv_nd(dims, channels, self.out_channels, eff_kernel, padding=eff_padding)
def forward(self, x, emb):
"""
Apply the block to a Tensor, conditioned on a timestep embedding.
:param x: an [N x C x ...] Tensor of features.
:param emb: an [N x emb_channels] Tensor of timestep embeddings.
:return: an [N x C x ...] Tensor of outputs.
"""
return checkpoint(
self._forward, x, emb
)
def _forward(self, x, emb):
h = self.in_layers(x)
emb_out = self.emb_layers(emb).type(h.dtype)
while len(emb_out.shape) < len(h.shape):
emb_out = emb_out[..., None]
if self.use_scale_shift_norm:
out_norm, out_rest = self.out_layers[0], self.out_layers[1:]
scale, shift = torch.chunk(emb_out, 2, dim=1)
h = out_norm(h) * (1 + scale) + shift
h = out_rest(h)
else:
h = h + emb_out
h = self.out_layers(h)
return self.skip_connection(x) + h
class DiffusionLayer(TimestepBlock):
def __init__(self, model_channels, dropout, num_heads):
super().__init__()
2022-05-20 23:04:22 +00:00
self.resblk = TimestepResBlock(model_channels, model_channels, dropout, model_channels, dims=1, use_scale_shift_norm=True)
2022-05-20 17:01:48 +00:00
self.attn = AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True)
def forward(self, x, time_emb):
y = self.resblk(x, time_emb)
return self.attn(y)
class FlatDiffusion(nn.Module):
def __init__(
self,
model_channels=512,
num_layers=8,
in_channels=256,
in_latent_channels=512,
in_vectors=8,
in_groups=8,
out_channels=512, # mean and variance
dropout=0,
use_fp16=False,
num_heads=8,
# Parameters for regularization.
layer_drop=.1,
unconditioned_percentage=.1, # This implements a mechanism similar to what is used in classifier-free training.
2022-05-23 05:10:58 +00:00
train_mel_head=False,
2022-05-20 17:01:48 +00:00
):
super().__init__()
self.in_channels = in_channels
self.model_channels = model_channels
self.out_channels = out_channels
self.dropout = dropout
self.num_heads = num_heads
self.unconditioned_percentage = unconditioned_percentage
self.enable_fp16 = use_fp16
self.layer_drop = layer_drop
self.inp_block = conv_nd(1, in_channels, model_channels, 3, 1, 1)
self.time_embed = nn.Sequential(
linear(model_channels, model_channels),
nn.SiLU(),
linear(model_channels, model_channels),
)
# Either code_converter or latent_converter is used, depending on what type of conditioning data is fed.
# This model is meant to be able to be trained on both for efficiency purposes - it is far less computationally
# complex to generate tokens, while generating latents will normally mean propagating through a deep autoregressive
# transformer network.
self.embeddings = nn.ModuleList([nn.Embedding(in_vectors, model_channels//in_groups) for _ in range(in_groups)])
self.latent_conditioner = nn.Sequential(
nn.Conv1d(in_latent_channels, model_channels, 3, padding=1),
2022-05-23 05:10:58 +00:00
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
2022-05-20 17:01:48 +00:00
)
2022-05-20 17:25:00 +00:00
self.code_converter = nn.Sequential(
2022-05-23 05:10:58 +00:00
ResBlock(dims=1, channels=model_channels, dropout=dropout),
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
ResBlock(dims=1, channels=model_channels, dropout=dropout),
AttentionBlock(model_channels, num_heads, relative_pos_embeddings=True),
ResBlock(dims=1, channels=model_channels, dropout=dropout),
2022-05-20 17:25:00 +00:00
)
2022-05-23 05:10:58 +00:00
self.code_norm = normalization(model_channels)
self.contextual_embedder = nn.Sequential(nn.Conv1d(in_channels,model_channels,3,padding=1,stride=2),
nn.Conv1d(model_channels, model_channels*2,3,padding=1,stride=2),
AttentionBlock(model_channels*2, num_heads, relative_pos_embeddings=True, do_checkpoint=False),
AttentionBlock(model_channels*2, num_heads, relative_pos_embeddings=True, do_checkpoint=False),
AttentionBlock(model_channels*2, num_heads, relative_pos_embeddings=True, do_checkpoint=False),
AttentionBlock(model_channels*2, num_heads, relative_pos_embeddings=True, do_checkpoint=False),
AttentionBlock(model_channels*2, num_heads, relative_pos_embeddings=True, do_checkpoint=False))
2022-05-20 17:01:48 +00:00
self.unconditioned_embedding = nn.Parameter(torch.randn(1,model_channels,1))
2022-05-23 05:10:58 +00:00
self.conditioning_timestep_integrator = TimestepEmbedSequential(
DiffusionLayer(model_channels, dropout, num_heads),
DiffusionLayer(model_channels, dropout, num_heads),
DiffusionLayer(model_channels, dropout, num_heads),
)
2022-05-20 17:01:48 +00:00
self.integrating_conv = nn.Conv1d(model_channels*2, model_channels, kernel_size=1)
self.mel_head = nn.Conv1d(model_channels, in_channels, kernel_size=3, padding=1)
self.layers = nn.ModuleList([DiffusionLayer(model_channels, dropout, num_heads) for _ in range(num_layers)] +
2022-05-20 23:04:22 +00:00
[TimestepResBlock(model_channels, model_channels, dropout, dims=1, use_scale_shift_norm=True) for _ in range(3)])
2022-05-20 17:01:48 +00:00
self.out = nn.Sequential(
normalization(model_channels),
nn.SiLU(),
zero_module(conv_nd(1, model_channels, out_channels, 3, padding=1)),
)
2022-05-23 05:10:58 +00:00
if train_mel_head:
for m in [self.conditioning_timestep_integrator, self.integrating_conv, self.layers,
self.out]:
for p in m.parameters():
p.requires_grad = False
p.DO_NOT_TRAIN = True
2022-05-20 23:04:22 +00:00
2022-05-20 17:01:48 +00:00
def get_grad_norm_parameter_groups(self):
groups = {
2022-05-23 05:10:58 +00:00
'minicoder': list(self.contextual_embedder.parameters()),
'layers': list(self.layers.parameters()),
2022-05-20 17:25:00 +00:00
'code_converters': list(self.embeddings.parameters()) + list(self.code_converter.parameters()) + list(self.latent_conditioner.parameters()),
2022-05-23 05:10:58 +00:00
'timestep_integrator': list(self.conditioning_timestep_integrator.parameters()) + list(self.integrating_conv.parameters()),
2022-05-20 17:01:48 +00:00
'time_embed': list(self.time_embed.parameters()),
}
return groups
2022-05-23 05:10:58 +00:00
def timestep_independent(self, aligned_conditioning, conditioning_input, expected_seq_len, return_code_pred):
# Shuffle aligned_latent to BxCxS format
if is_latent(aligned_conditioning):
aligned_conditioning = aligned_conditioning.permute(0, 2, 1)
2022-05-20 17:01:48 +00:00
2022-05-23 05:10:58 +00:00
# Note: this block does not need to repeated on inference, since it is not timestep-dependent or x-dependent.
speech_conditioning_input = conditioning_input.unsqueeze(1) if len(
conditioning_input.shape) == 3 else conditioning_input
conds = []
for j in range(speech_conditioning_input.shape[1]):
conds.append(self.contextual_embedder(speech_conditioning_input[:, j]))
conds = torch.cat(conds, dim=-1)
cond_emb = conds.mean(dim=-1)
cond_scale, cond_shift = torch.chunk(cond_emb, 2, dim=1)
if is_latent(aligned_conditioning):
code_emb = self.latent_conditioner(aligned_conditioning)
else:
code_emb = [embedding(aligned_conditioning[:, :, i]) for i, embedding in enumerate(self.embeddings)]
code_emb = torch.cat(code_emb, dim=-1).permute(0,2,1)
2022-05-20 17:01:48 +00:00
unconditioned_batches = torch.zeros((code_emb.shape[0], 1, 1), device=code_emb.device)
# Mask out the conditioning branch for whole batch elements, implementing something similar to classifier-free guidance.
if self.training and self.unconditioned_percentage > 0:
unconditioned_batches = torch.rand((code_emb.shape[0], 1, 1),
device=code_emb.device) < self.unconditioned_percentage
2022-05-23 05:10:58 +00:00
code_emb = torch.where(unconditioned_batches, self.unconditioned_embedding.repeat(aligned_conditioning.shape[0], 1, 1),
2022-05-20 17:01:48 +00:00
code_emb)
expanded_code_emb = F.interpolate(code_emb, size=expected_seq_len, mode='nearest')
2022-05-23 05:10:58 +00:00
expanded_code_emb = self.code_converter(expanded_code_emb)
expanded_code_emb = self.code_norm(expanded_code_emb) * (1 + cond_scale.unsqueeze(-1)) + cond_shift.unsqueeze(-1)
2022-05-20 17:01:48 +00:00
if not return_code_pred:
2022-05-23 05:10:58 +00:00
return expanded_code_emb
2022-05-20 17:01:48 +00:00
else:
2022-05-23 05:10:58 +00:00
mel_pred = self.mel_head(expanded_code_emb)
# Multiply mel_pred by !unconditioned_branches, which drops the gradient on unconditioned branches. This is because we don't want that gradient being used to train parameters through the codes_embedder as it unbalances contributions to that network from the MSE loss.
2022-05-20 17:01:48 +00:00
mel_pred = mel_pred * unconditioned_batches.logical_not()
2022-05-23 05:10:58 +00:00
return expanded_code_emb, mel_pred
2022-05-20 17:01:48 +00:00
2022-05-23 05:10:58 +00:00
def forward(self, x, timesteps, aligned_conditioning=None, conditioning_input=None, precomputed_aligned_embeddings=None, conditioning_free=False, return_code_pred=False):
2022-05-20 17:01:48 +00:00
"""
Apply the model to an input batch.
:param x: an [N x C x ...] Tensor of inputs.
:param timesteps: a 1-D batch of timesteps.
2022-05-23 05:10:58 +00:00
:param aligned_conditioning: an aligned latent or sequence of tokens providing useful data about the sample to be produced.
2022-05-20 17:01:48 +00:00
:param conditioning_input: a full-resolution audio clip that is used as a reference to the style you want decoded.
2022-05-23 05:10:58 +00:00
:param precomputed_aligned_embeddings: Embeddings returned from self.timestep_independent()
2022-05-20 17:01:48 +00:00
:param conditioning_free: When set, all conditioning inputs (including tokens and conditioning_input) will not be considered.
:return: an [N x C x ...] Tensor of outputs.
"""
2022-05-23 05:10:58 +00:00
assert precomputed_aligned_embeddings is not None or (aligned_conditioning is not None and conditioning_input is not None)
assert not (return_code_pred and precomputed_aligned_embeddings is not None) # These two are mutually exclusive.
2022-05-20 17:01:48 +00:00
unused_params = []
if conditioning_free:
code_emb = self.unconditioned_embedding.repeat(x.shape[0], 1, x.shape[-1])
unused_params.extend(list(self.code_converter.parameters()) + list(self.code_embedding.parameters()))
unused_params.extend(list(self.latent_conditioner.parameters()))
else:
2022-05-23 05:10:58 +00:00
if precomputed_aligned_embeddings is not None:
code_emb = precomputed_aligned_embeddings
2022-05-20 17:01:48 +00:00
else:
2022-05-23 05:10:58 +00:00
code_emb, mel_pred = self.timestep_independent(aligned_conditioning, conditioning_input, x.shape[-1], True)
if is_latent(aligned_conditioning):
unused_params.extend(list(self.code_converter.parameters()) + list(self.code_embedding.parameters()))
else:
unused_params.extend(list(self.latent_conditioner.parameters()))
2022-05-20 17:01:48 +00:00
unused_params.append(self.unconditioned_embedding)
2022-05-23 05:10:58 +00:00
time_emb = self.time_embed(timestep_embedding(timesteps, self.model_channels))
code_emb = self.conditioning_timestep_integrator(code_emb, time_emb)
2022-05-20 17:01:48 +00:00
x = self.inp_block(x)
x = torch.cat([x, code_emb], dim=1)
x = self.integrating_conv(x)
for i, lyr in enumerate(self.layers):
# Do layer drop where applicable. Do not drop first and last layers.
if self.training and self.layer_drop > 0 and i != 0 and i != (len(self.layers)-1) and random.random() < self.layer_drop:
unused_params.extend(list(lyr.parameters()))
else:
# First and last blocks will have autocast disabled for improved precision.
with autocast(x.device.type, enabled=self.enable_fp16 and i != 0):
2022-05-23 05:10:58 +00:00
x = lyr(x, time_emb)
2022-05-20 17:01:48 +00:00
x = x.float()
out = self.out(x)
# Involve probabilistic or possibly unused parameters in loss so we don't get DDP errors.
extraneous_addition = 0
for p in unused_params:
extraneous_addition = extraneous_addition + p.mean()
out = out + extraneous_addition * 0
if return_code_pred:
return out, mel_pred
return out
def get_conditioning_latent(self, conditioning_input):
speech_conditioning_input = conditioning_input.unsqueeze(1) if len(
conditioning_input.shape) == 3 else conditioning_input
conds = []
for j in range(speech_conditioning_input.shape[1]):
2022-05-23 05:10:58 +00:00
conds.append(self.contextual_embedder(speech_conditioning_input[:, j]))
2022-05-20 17:01:48 +00:00
conds = torch.cat(conds, dim=-1)
return conds.mean(dim=-1)
@register_model
def register_flat_diffusion(opt_net, opt):
return FlatDiffusion(**opt_net['kwargs'])
if __name__ == '__main__':
clip = torch.randn(2, 256, 400)
2022-05-23 05:10:58 +00:00
aligned_latent = torch.randn(2,388,512)
2022-05-20 17:01:48 +00:00
aligned_sequence = torch.randint(0,8,(2,100,8))
cond = torch.randn(2, 256, 400)
ts = torch.LongTensor([600, 600])
2022-05-23 05:10:58 +00:00
model = FlatDiffusion(512, layer_drop=.3, unconditioned_percentage=.5, train_mel_head=True)
# Test with latent aligned conditioning
#o = model(clip, ts, aligned_latent, cond)
# Test with sequence aligned conditioning
2022-05-20 17:25:00 +00:00
o = model(clip, ts, aligned_sequence, cond, return_code_pred=True)
2022-05-20 17:01:48 +00:00