mrq/vall-e
5
9
Fork 0
Code Issues 8 Pull Requests Packages Projects Releases Wiki Activity

small slop cleanup

Browse Source
This commit is contained in:
mrq 2025-02-24 19:03:53 -06:00
parent 3330b5bb00
commit 918e0dbac1
1 changed files with 5 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
vall_e/models/base.py
View File

@ -371,14 +371,12 @@ class AudioEncoder(nn.Module):
""" """
# encode by interleaving # encode by interleaving
# resultant tensor is equal to prior naive attempt
seq_len = xi.shape[0] seq_len = xi.shape[0]
# (8, seq_len, dim) # (seq_len, 8, dim)
x = [ emb( xi[:, l] ) for l, emb in enumerate(self.embs) ] x = torch.stack([emb(xi[:, l]) for l, emb in enumerate(self.embs)], dim=1)
# => (seq_len, dim * 8) interleaved # (seq_len, 8 * dim)
x_i = [] x = x.view(x.shape[0], -1)
for i in range(xi.shape[0]):
x_i.append(torch.cat([ x[l][i] for l in range(len(self.embs)) ], dim=-1))
x = torch.stack( x_i, dim=0 )
# => (seq_len, dim) # => (seq_len, dim)
x = self.proj(x) x = self.proj(x)