From 7aa168585b1f784f071ae555c869e22bfec762ad Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Tue, 3 Jan 2023 13:45:10 +0900
Subject: [PATCH] from local

---
 .../.data2vec_audio.py.swp                    | Bin 45056 -> 45056 bytes
 .../data2vec_audio.py                         |   6 +++---
 2 files changed, 3 insertions(+), 3 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 73163ab5fcfe8555f4a2185a419cd70adebc7e44..2c2c31722dbbae7bd9ca3f9914e39206b3b322a7 100644
GIT binary patch
delta 399
zcmW;8JxBrp0EXc&<*Z9Baj?cS>l!Y?rJ*5RqMxCmf*Ohh566ab3{nVO`*H5SpsC0s
z$c7r4!a}H}(XEJRFA{p`gXiNdHhjf~FP2TtO-@<EgGsT7G*t0iBX=45+BDAnl3g~p
zoWe(kbc#hRAc~i;bb&n5h{M8XNP56Mj&X!FEF*(a44?~Pgzy%Ws<=i8OBlx(te^^N
zwoAWw!U=MiM*`2LbbuYWh(fq)lWwt#0@mST2GfY6-YRV%gHAL9(j#g(LmsPvv@}bM
zpdU6u_%Wn!yy6B`9HImlS)^fmE8AE8N>9pcNt9F0#Q$66`d%+pbNamQ{rAWp{bWN@

delta 380
zcmWN@JxBrp07vmZCt8<v8HnT-C&5J`u1(?+IW(0)L(!E-V+1*d*ATc>)GqYl(3Bo6
z*<gcHR0O#+ITnGm7ukF8gXi}Qzh@*ivnz$gskuy66qC};V5?hdB>G3;n~>z3n#Z%1
zFH5Rp9czfAJ0>+yMjk2H=#NT|=->*M*vAfB%p-#&#$n+zDz$KjDmGBS0_><N8jMJP
zctIW8Sj9BnOz9kF@DPWAhheFW8V=yY!!nkT!t0Q92p1C=M5Jdt;TmP^MRHP+$zcW#
kEc}M0AH3rpEnJ`q4<+Q`1iL4<A;V2K?Qb)ftbN4(0nMU6T>t<8

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 335779d40..51a681b65 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
@@ -41,9 +41,9 @@ class TransformerEncoderAdapter(TransformerEncoder):
         super().__init__(args)
         self.adapters = ResidualAdapterModule()
 
-        for p in self.adapters.parameters():
-            #p.data = nn.Parameter(torch.zeros(p.size()).to('cuda'))
-            p.data = nn.Parameter(torch.randn(p.size()).to('cuda')/50.)
+        #for p in self.adapters.parameters():
+        #    p.data = nn.Parameter(torch.zeros(p.size()).to('cuda'))
+        #    p.data = nn.Parameter(torch.randn(p.size()).to('cuda')/50.)
 
     def forward(self, x, padding_mask=None, layer=None, tgt_layer=None):
         x, layer_results = self.extract_features_with_adapter(