mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-12-11 06:55:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'scaled_adam_exp465' into scaled_adam_exp472

Browse Source
This commit is contained in:
Daniel Povey 2022-11-23 14:16:17 +08:00
commit f89a85aed8
1 changed files with 0 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
View File

@ -1223,15 +1223,6 @@ class SelfAttention(nn.Module):
num_heads * value_head_dim,
bias=True)
# attempt to make the output of `in_proj` uncorrelated within each head
# and all heads having roughly the same magnitude. the hope is to
# improve learning dynamics; this loses no power as there is no constraint
# on the condition number of out_proj.
self.whiten_values = Whiten(num_groups=num_heads,
whitening_limit=2.0,
prob=(0.025, 0.25),
grad_scale=0.025)
self.out_proj = ScaledLinear(num_heads * value_head_dim,
embed_dim, bias=True,
initial_scale=0.05)
@ -1264,7 +1255,6 @@ class SelfAttention(nn.Module):
assert attn_weights.shape == (num_heads, batch_size, seq_len, seq_len)
x = self.in_proj(x) # (seq_len, batch_size, num_heads * value_head_dim)
x = self.whiten_values(x) # does nothing in the forward pass.
x = x.reshape(seq_len, batch_size, num_heads, -1).permute(2, 1, 0, 3)
# now x: (num_heads, batch_size, seq_len, value_head_dim)
value_head_dim = x.shape[-1]