From 3b1ca50e6fc882b36983f6f51f8764964dacc4c9 Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Tue, 11 Apr 2023 15:03:06 +0900
Subject: [PATCH] from local

---
 .../.data2vec_audio.py.swp                    | Bin 40960 -> 40960 bytes
 .../data2vec_audio.py                         |  14 +++++++-------
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.data2vec_audio.py.swp
index fdca8c5acef67c905d9e33c0d3384acbce448d8d..9d4f01465d38cb18a385ddd3fe2afc15cab6cb32 100644
GIT binary patch
delta 204
zcmXZL!3x1}00-dTM#~KCvgEdMw6ueWQk&GMIjxj(l6P=1=E`vo;ednwj?|t)x$yul
zE=mr52VZ?rB1egAOzdHQU^c5ZNC&>}+VplbX3p}G1I`?AhZR0)fZz?c*y0L5v%rC4
zUa`e#eyPBn*F5Jj2VCbOKQh3ccf8<`M{HyeYi?d@h>}}S;;yHw=3iMjUR#QF|0=@X
KHYGm368r#p&?evj

delta 211
zcmXZLtqy`<7zWT6oDM-92Ig=Ek`ZyU82lkAn-raFR!20lnV2MQ;M<Hw119%rWTRRw
z7B%7}Sbc(LJ5eS^nW(KT&vA{uX%VPI^KZ`&&)RmpyH$u13~+=)tfPt_g($}aH)!Eo
zCVJr&ee`gN4SbY|QaquDCbm$<i$rw97(-m4CJ|HYs&MDJT`RHYLFePiK#+_-)@|!N
RaN+{%xJ9;~raYfD7e5udDRBS*

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
index 8cae787ab..20adb7724 100644
--- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/data2vec_audio.py
@@ -465,13 +465,6 @@ class Data2VecAudioModel(BaseFairseqModel):
 
         features = features.transpose(1, 2)
 
-        ## for prompt tuning
-        if prompt is not None:
-            #features = torch.cat([features, prompt])
-            features = torch.cat([prompt, features])
-
-        features = self.layer_norm(features)
-
         orig_padding_mask = padding_mask
 
         if padding_mask is not None and padding_mask.any():
@@ -495,6 +488,13 @@ class Data2VecAudioModel(BaseFairseqModel):
         else:
             padding_mask = None
         
+        ## for prompt tuning
+        if prompt is not None:
+            #features = torch.cat([features, prompt])
+            features = torch.cat([prompt, features])
+
+        features = self.layer_norm(features)
+
         #print(padding_mask.size())
         #print((padding_mask[0] == True).nonzero(as_tuple=True)[0])
         #print((padding_mask[1] == True).nonzero(as_tuple=True)[0][1])