deviandice/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

iq checkin

Browse Source
This commit is contained in:
James Betker 2022-07-20 10:19:32 -06:00
parent 9a37f3ba42
commit 7b3fc79737
1 changed files with 10 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
codes/models/audio/music/instrument_quantizer.py
View File

@ -28,7 +28,6 @@ class SelfClassifyingHead(nn.Module):
h = self.dec(x) h = self.dec(x)
o = self.to_output(h[:, -1]) o = self.to_output(h[:, -1])
q, c, _ = self.quantizer(o, used_codes) q, c, _ = self.quantizer(o, used_codes)
q = torch.sigmoid(q)
return q, c return q, c
def forward(self, x, target): def forward(self, x, target):
@ -37,10 +36,13 @@ class SelfClassifyingHead(nn.Module):
outputs = [] outputs = []
results = [] results = []
codes = [] codes = []
q_reg = 0
for i in range(self.seq_len): for i in range(self.seq_len):
q, c = checkpoint(functools.partial(self.do_ar_step, used_codes=codes), torch.stack(stack, dim=1)) q, c = checkpoint(functools.partial(self.do_ar_step, used_codes=codes), torch.stack(stack, dim=1))
q_reg = q_reg + (q ** 2).mean()
s = torch.sigmoid(q)
outputs.append(q) outputs.append(s)
output = torch.stack(outputs, dim=1).sum(1) output = torch.stack(outputs, dim=1).sum(1)
# If the addition would strictly make the result worse, set it to 0. Sometimes. # If the addition would strictly make the result worse, set it to 0. Sometimes.
@ -49,13 +51,13 @@ class SelfClassifyingHead(nn.Module):
probabilistic_worsen = torch.rand_like(worsen) * worsen > .5 probabilistic_worsen = torch.rand_like(worsen) * worsen > .5
output = output * probabilistic_worsen.unsqueeze(-1) # This is non-differentiable, but still deterministic. output = output * probabilistic_worsen.unsqueeze(-1) # This is non-differentiable, but still deterministic.
c[probabilistic_worsen] = -1 # Code of -1 means the code was unused. c[probabilistic_worsen] = -1 # Code of -1 means the code was unused.
q = q * probabilistic_worsen.unsqueeze(-1) s = s * probabilistic_worsen.unsqueeze(-1)
outputs[-1] = q outputs[-1] = s
codes.append(c) codes.append(c)
stack.append(self.to_decoder(q)) stack.append(self.to_decoder(s))
results.append(output) results.append(output)
return results, torch.cat(codes, dim=0) return results, torch.cat(codes, dim=0), q_reg / self.seq_len
class VectorResBlock(nn.Module): class VectorResBlock(nn.Module):
@ -112,13 +114,13 @@ class InstrumentQuantizer(nn.Module):
for lyr in self.encoder: for lyr in self.encoder:
h = lyr(h) h = lyr(h)
reconstructions, codes = self.heads(h, f) reconstructions, codes, q_reg = self.heads(h, f)
reconstruction_losses = torch.stack([F.mse_loss(r.reshape(b, s, c), px) for r in reconstructions]) reconstruction_losses = torch.stack([F.mse_loss(r.reshape(b, s, c), px) for r in reconstructions])
r_follow = torch.arange(1, reconstruction_losses.shape[0]+1, device=x.device) r_follow = torch.arange(1, reconstruction_losses.shape[0]+1, device=x.device)
reconstruction_losses = (reconstruction_losses * r_follow / r_follow.shape[0]) reconstruction_losses = (reconstruction_losses * r_follow / r_follow.shape[0])
self.log_codes(codes) self.log_codes(codes)
return reconstruction_losses return reconstruction_losses, q_reg
def log_codes(self, codes): def log_codes(self, codes):
if self.internal_step % 5 == 0: if self.internal_step % 5 == 0: