blakelemoine/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Pull Requests Activity

.

Browse Source
This commit is contained in:
James Betker 2022-06-10 15:35:36 -06:00
parent ee2827dee9
commit dca16e6447
1 changed files with 28 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

57
codes/models/audio/music/transformer_diffusion8.py
View File

@ -71,6 +71,7 @@ class TransformerDiffusion(nn.Module):
rotary_emb_dim=32, rotary_emb_dim=32,
input_vec_dim=512, input_vec_dim=512,
out_channels=512, # mean and variance out_channels=512, # mean and variance
num_heads=16,
dropout=0, dropout=0,
use_fp16=False, use_fp16=False,
ar_prior=False, ar_prior=False,
@ -94,7 +95,6 @@ class TransformerDiffusion(nn.Module):
nn.SiLU(), nn.SiLU(),
linear(prenet_channels, model_channels), linear(prenet_channels, model_channels),
) )
prenet_heads = prenet_channels//64
self.ar_prior = ar_prior self.ar_prior = ar_prior
if ar_prior: if ar_prior:
@ -102,7 +102,7 @@ class TransformerDiffusion(nn.Module):
self.ar_prior_intg = Encoder( self.ar_prior_intg = Encoder(
dim=prenet_channels, dim=prenet_channels,
depth=prenet_layers, depth=prenet_layers,
heads=prenet_heads, heads=num_heads,
ff_dropout=dropout, ff_dropout=dropout,
attn_dropout=dropout, attn_dropout=dropout,
use_rmsnorm=True, use_rmsnorm=True,
@ -116,7 +116,7 @@ class TransformerDiffusion(nn.Module):
self.code_converter = Encoder( self.code_converter = Encoder(
dim=prenet_channels, dim=prenet_channels,
depth=prenet_layers, depth=prenet_layers,
heads=prenet_heads, heads=num_heads,
ff_dropout=dropout, ff_dropout=dropout,
attn_dropout=dropout, attn_dropout=dropout,
use_rmsnorm=True, use_rmsnorm=True,
@ -129,7 +129,7 @@ class TransformerDiffusion(nn.Module):
self.unconditioned_embedding = nn.Parameter(torch.randn(1,1,prenet_channels)) self.unconditioned_embedding = nn.Parameter(torch.randn(1,1,prenet_channels))
self.rotary_embeddings = RotaryEmbedding(rotary_emb_dim) self.rotary_embeddings = RotaryEmbedding(rotary_emb_dim)
self.intg = nn.Linear(prenet_channels*2, model_channels) self.intg = nn.Linear(prenet_channels*2, model_channels)
self.layers = TimestepRotaryEmbedSequential(*[DietAttentionBlock(model_channels, block_channels, block_channels // 64, dropout) for _ in range(num_layers)]) self.layers = TimestepRotaryEmbedSequential(*[DietAttentionBlock(model_channels, block_channels, num_heads, dropout) for _ in range(num_layers)])
self.out = nn.Sequential( self.out = nn.Sequential(
normalization(model_channels), normalization(model_channels),
@ -196,18 +196,16 @@ class TransformerDiffusion(nn.Module):
class TransformerDiffusionWithQuantizer(nn.Module): class TransformerDiffusionWithQuantizer(nn.Module):
def __init__(self, freeze_quantizer_until=20000, quantizer_dims=[1024], no_reconstruction=True, **kwargs): def __init__(self, quantizer_dims=[1024], freeze_quantizer_until=20000, **kwargs):
super().__init__() super().__init__()
self.internal_step = 0 self.internal_step = 0
self.freeze_quantizer_until = freeze_quantizer_until self.freeze_quantizer_until = freeze_quantizer_until
self.diff = TransformerDiffusion(**kwargs) self.diff = TransformerDiffusion(**kwargs)
self.quantizer = MusicQuantizer2(inp_channels=kwargs['in_channels'], inner_dim=quantizer_dims, self.quantizer = MusicQuantizer2(inp_channels=kwargs['in_channels'], inner_dim=quantizer_dims,
codevector_dim=quantizer_dims[0], checkpoint=False, codevector_dim=quantizer_dims[0], codebook_size=256,
codebook_size=256, codebook_groups=2, codebook_groups=2, max_gumbel_temperature=4, min_gumbel_temperature=.5)
max_gumbel_temperature=4, min_gumbel_temperature=.5)
self.quantizer.quantizer.temperature = self.quantizer.min_gumbel_temperature self.quantizer.quantizer.temperature = self.quantizer.min_gumbel_temperature
if no_reconstruction:
del self.quantizer.up del self.quantizer.up
def update_for_step(self, step, *args): def update_for_step(self, step, *args):
@ -219,30 +217,27 @@ class TransformerDiffusionWithQuantizer(nn.Module):
) )
def forward(self, x, timesteps, truth_mel, conditioning_input=None, disable_diversity=False, conditioning_free=False): def forward(self, x, timesteps, truth_mel, conditioning_input=None, disable_diversity=False, conditioning_free=False):
mse, diversity_loss, proj = self.quantizer(truth_mel, return_decoder_latent=True) quant_grad_enabled = self.internal_step > self.freeze_quantizer_until
with torch.set_grad_enabled(quant_grad_enabled):
proj, diversity_loss = self.quantizer(truth_mel, return_decoder_latent=True)
proj = proj.permute(0,2,1) proj = proj.permute(0,2,1)
quant_grad_enabled = self.internal_step > self.freeze_quantizer_until
if not quant_grad_enabled:
proj = proj.detach()
# Make sure this does not cause issues in DDP by explicitly using the parameters for nothing. # Make sure this does not cause issues in DDP by explicitly using the parameters for nothing.
if not quant_grad_enabled:
unused = 0 unused = 0
for p in self.quantizer.parameters(): for p in self.quantizer.parameters():
unused = unused + p.mean() * 0 unused = unused + p.mean() * 0
proj = proj + unused proj = proj + unused
diversity_loss = diversity_loss * 0
diff = self.diff(x, timesteps, codes=proj, conditioning_input=conditioning_input, diff = self.diff(x, timesteps, codes=proj, conditioning_input=conditioning_input, conditioning_free=conditioning_free)
conditioning_free=conditioning_free)
if disable_diversity: if disable_diversity:
return diff return diff
if mse is None:
return diff, diversity_loss return diff, diversity_loss
return diff, diversity_loss, mse
def get_debug_values(self, step, __): def get_debug_values(self, step, __):
if self.quantizer.total_codes > 0: if self.quantizer.total_codes > 0:
return {'histogram_quant_codes': self.quantizer.codes[:self.quantizer.total_codes], return {'histogram_codes': self.quantizer.codes[:self.quantizer.total_codes],
'gumbel_temperature': self.quantizer.quantizer.temperature} 'gumbel_temperature': self.quantizer.quantizer.temperature}
else: else:
return {} return {}
@ -320,26 +315,24 @@ def register_transformer_diffusion8_with_ar_prior(opt_net, opt):
def test_quant_model(): def test_quant_model():
clip = torch.randn(2, 100, 401) clip = torch.randn(2, 256, 400)
cond = torch.randn(2, 256, 400)
ts = torch.LongTensor([600, 600]) ts = torch.LongTensor([600, 600])
model = TransformerDiffusionWithQuantizer(in_channels=100, out_channels=200, quantizer_dims=[1024,768,512,384], model = TransformerDiffusionWithQuantizer(in_channels=256, model_channels=2048, block_channels=1024,
model_channels=2048, block_channels=1024, prenet_channels=1024, prenet_channels=1024, num_heads=8,
input_vec_dim=1024, num_layers=16, prenet_layers=6, input_vec_dim=1024, num_layers=16, prenet_layers=6)
no_reconstruction=False) model.get_grad_norm_parameter_groups()
#model.get_grad_norm_parameter_groups()
#quant_weights = torch.load('D:\\dlas\\experiments\\train_music_quant_r4\\models\\5000_generator.pth') quant_weights = torch.load('D:\\dlas\\experiments\\train_music_quant_r4\\models\\5000_generator.pth')
#diff_weights = torch.load('X:\\dlas\\experiments\\train_music_diffusion_tfd5\\models\\48000_generator_ema.pth') model.quantizer.load_state_dict(quant_weights, strict=False)
#model.quantizer.load_state_dict(quant_weights, strict=False)
#model.diff.load_state_dict(diff_weights)
#torch.save(model.state_dict(), 'sample.pth') torch.save(model.state_dict(), 'sample.pth')
print_network(model) print_network(model)
o = model(clip, ts, clip) o = model(clip, ts, clip, cond)
def test_ar_model(): def test_ar_model():
clip = torch.randn(2, 256, 401) clip = torch.randn(2, 256, 400)
cond = torch.randn(2, 256, 400) cond = torch.randn(2, 256, 400)
ts = torch.LongTensor([600, 600]) ts = torch.LongTensor([600, 600])
model = TransformerDiffusionWithARPrior(model_channels=2048, block_channels=1024, prenet_channels=1024, model = TransformerDiffusionWithARPrior(model_channels=2048, block_channels=1024, prenet_channels=1024,