From 2a63e6bcc3f35d84bcd4156dca9ed31e9cfc76ed Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Sat, 10 Dec 2022 14:12:58 +0900
Subject: [PATCH] from local

---
 .../.train.py.swp                             | Bin 106496 -> 106496 bytes
 .../train.py                                  |  31 +++++++++---------
 2 files changed, 15 insertions(+), 16 deletions(-)

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train.py.swp b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/.train.py.swp
index 7ccb440a8bac8b01e7cbe09cf6f24a5ac24279f7..341ef94a3f0955a139a005ddd4e1307af06b9134 100644
GIT binary patch
delta 1253
zcmY+^Ye<t(9LMo<Y|iGII`Wd3;U!8-%xXramQm^|LKl(d3+E*b!8B_j5wVJbE-#us
z)?-211-cPMjanwuN-9zeEh`ArK#fo#A}BGte4p}#9r&D`^WS-%hwcC4IpEbF@ajWK
zlD4ed7?qNkBp5+LSVJ00y|+Vt1j!u5Yz{MoYjsAbVq<6ew-M!To7w%)IqV}ula3k}
zmQQ~5J5Y#Gj9?hI(1=}Fk2w4a5Mm5N7{masp#gbV1Oo<mUN1V(ju$wAQly%MQ-rVx
z#>5N`qaAMSgcUKEFbeSvLm0#W9^*1xsK#>m;gi1*&+rucu?(XIAzq>rPGle)vkXol
z0$2!iO&6jKS8y8n$i^I)Fz6@5J2b+D8q9|tkMu&^#RZ&4DN3*s2?)cuFaHF*cz_n%
zNMLfofvs4HSVY1fA8E}Ho}tNOv&k<$>dwtlr+Uk!&OO#$V7&7|muqrO<A+anF~?(0
zlB;C2w(rC#97jH4@sqPXX!X=AmN`DbQBhIF4o6v)qoTUXzO7<Mx}(~z-d`%?wP~I<
zC2P|XBM1KuPHTnb5mO3NFW1USeP6uCoGiZ<nDS^!2egERs%Dprb=DCK2eJ`~LHd6a
z&Cr4|7jNnNRg@tTZzgZ^gU_$HL~Sp2qX>2sVlD#liZj=77=_437BY>5ZY>Kv`UuB8
zxNs7+C`2X_F+wida1OPoK@sdIL>i)D#4s6g;}A-*8W|WPAGfg`>tOXGbmIh~6%9~u
z4EwMFT4-Wm#%nV01ZQv*c5K2-7%@siFXA9pAPKV(fj*ktixyl#J@z0!ohcQ`uwXun
z_{_VY2QAt=0gX6<ViaLD!hPzx7ycK)>jtIkOsLY?w^lLrTNPXX8pYM0s66P;(rHQ1
zqOwN)zEr09Y1{Y&q2wx|s?{Rn)Jls?Q7>5J0`<9tyuELh0qVF#2Kts&?k-#5Iol-5
F<sUSt=Ys$M

delta 1317
zcmXxjdu&Tl9Ki8&*n_n*CCiGjn#WR;Ju+*J%~&KJ*=Ck8OjNgaT(gJT5|(JYh)0BE
z6ep-P1o4VjNBILYuf#thM6!rPCLR%g2>$Sh#rImB<a5tC_i^q$zu)ird*b{(amlqg
z%NLbd3bJwpZKM#hk`LA$zMA}dq^u5@Oexkx{XA{hU?=+I;2_$GG12pl;iabV9pCa1
zLL7j;)c;3rBnhz>Wf+c!T)&G*^xli~)v!b2B=2;h5=$`$MJR;CC!-J#5JWpFuo!lu
z@Ch4(eEc>D@gC>Ufg0o^8{?6Rad@2|#C7b$I@F^a`A9`9dWQ>f6s;)7LfBzLK1SmQ
zyX_z56XG6&Jy?lZFe3#|<AvCbIj|rN&*F#{8n6LtQHgXU<7cc8A8{T>&;kc+u;Nn;
zFU5Uq#R?w>6XNl0s1RRq9o^`J7aB^DjZrYdfPo<_hDW%7PPCyJt1%OHnDHf=eoM4z
z^_KLyHjg8G=|!bMg}P*#+39j=ty*)FbEBhnV{=_YQI)67J>4=%J@1m%!y#!l<S=V-
zdYp}$vzVtSTS!h(j*v9Q*=;lRI~ziBvI>NxB~xo^X=`<C*4o`ggV&<Zxtr>mU2c!V
zr8Q0$`nR~7)W;dpnvju^@o#Pws_&RARo0-KuQml`vFZ-WX{tXcORl|%sYsm5W^S^1
zy<V#$rW{!;C&m)l&je79>T#@x4Vln`dNZ;qS1uW%VrR$+s{FVdAO3akgCWviEYm9z
zH6l|GU9eG^I^`taQEGY@c47w>BM*sqM?K%i5d`3e4_?eiHoj8MPjCueY(YKTaA7J2
zDCL*9g3D0w!foK6C}&{6TWa|>6!yapAG}zNdB{Wpo>S8&u^mpVg&EH%<z3LQ9%J!}
zQtn0==Wq-SSc8&y{)t=$->Bz)+`?J-;6WZTVMH&HKY>-KKp|#hfXM$rA0Fcx4q*#w
zunbjL06j}d_)Ze~aL25t45!fnKb$Dg)3{QqmMR&edg`Sy@FgoLeCEX}Lm+RPDRQw(
GR>{Aa!r*TJ

diff --git a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
index ccd949a7c..9a23d9a1b 100755
--- a/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
+++ b/egs/librispeech/ASR/pruned_transducer_stateless_d2v_v2/train.py
@@ -534,22 +534,6 @@ def get_encoder_model(params: AttributeDict) -> nn.Module:
     def to_int_tuple(s: str):
         return tuple(map(int, s.split(",")))
     
-    '''
-    encoder = Zipformer(
-        num_features=params.feature_dim,
-        output_downsampling_factor=2,
-        zipformer_downsampling_factors=to_int_tuple(
-            params.zipformer_downsampling_factors
-        ),
-        encoder_dims=to_int_tuple(params.encoder_dims),
-        attention_dim=to_int_tuple(params.attention_dims),
-        encoder_unmasked_dims=to_int_tuple(params.encoder_unmasked_dims),
-        nhead=to_int_tuple(params.nhead),
-        feedforward_dim=to_int_tuple(params.feedforward_dims),
-        cnn_module_kernels=to_int_tuple(params.cnn_module_kernels),
-        num_encoder_layers=to_int_tuple(params.num_encoder_layers),
-    )
-    '''
     if params.encoder_type == 'd2v':
         encoder = FairSeqData2VecEncoder(
                     input_size=params.encoder_dim,
@@ -558,6 +542,21 @@ def get_encoder_model(params: AttributeDict) -> nn.Module:
                     freeze_finetune_updates=params.freeze_finetune_updates*params.accum_grads,
                     additional_block=params.additional_block,
                 ) 
+    else:
+        encoder = Zipformer(
+            num_features=params.feature_dim,
+            output_downsampling_factor=2,
+            zipformer_downsampling_factors=to_int_tuple(
+                params.zipformer_downsampling_factors
+            ),
+            encoder_dims=to_int_tuple(params.encoder_dims),
+            attention_dim=to_int_tuple(params.attention_dims),
+            encoder_unmasked_dims=to_int_tuple(params.encoder_unmasked_dims),
+            nhead=to_int_tuple(params.nhead),
+            feedforward_dim=to_int_tuple(params.feedforward_dims),
+            cnn_module_kernels=to_int_tuple(params.cnn_module_kernels),
+            num_encoder_layers=to_int_tuple(params.num_encoder_layers),
+        )
 
     return encoder