From d61e27625f0d0af0b7a6b3062006057e368eddb7 Mon Sep 17 00:00:00 2001
From: dohe0342 <kimdohe1070@gmail.com>
Date: Mon, 9 Jan 2023 19:43:22 +0900
Subject: [PATCH] from local

---
 .../ASR/incremental_transf/.conformer.py.swp  | Bin 114688 -> 114688 bytes
 .../ASR/incremental_transf/conformer.py       |  64 ++----------------
 2 files changed, 4 insertions(+), 60 deletions(-)

diff --git a/egs/librispeech/ASR/incremental_transf/.conformer.py.swp b/egs/librispeech/ASR/incremental_transf/.conformer.py.swp
index d8c62d0ccb038f352c3ef26127beaeb87cb0070a..c70f2b9218662a177e50d0ad37152e0df086fd7b 100644
GIT binary patch
delta 1201
zcmZwFUr19?90%~<QRj7c^Uu|oq|RN<=|-ilEK(_<g0QHF$d{0sdIe!_akt^3Xp4N3
z406QrA+$f(OHmUa7R0c(2&9n6-XbuBh|C@`2+F=^|5EF~=iJ}<o!>e4--$$|NJO&q
zG<S6#S6dsK39W?4^YvNp+^=cN&k|;eDvFJDTyrULm%`rM2){QFt>&ldIz&C%S~1aY
zSb$d$gnF<;CCnEQT?HrnC?uKz4f4SNt7f9dFbrMb02ORd2x}&yPw>`6VTvIb0}mVl
zCse>knP?H7LLA0{!yz~b<*;bP&0!pdpb6|y2j2>a5-<unoB|`PNq9MU08>(!D1=}L
zYC(L7@9+*@!fgn{6=;JTV6cR*_5!YhAI`&Wu!9Vr^N3<_3ogR|9E1H(1xvX^voL8!
z7XUvr#YgY5dyHG}v%^dkm+hiUMhzm`E4rE@HQoH0r@ygB<f_Xy=G-2)w^vMT6<ww1
z+C^te4*#=3i&;(WwsGaUMXATwCpJhwoK?gvQhK2nR5LY>i)>WfefYRI9bZ74KoeJD
z%oe_c!*c*kAcL5WfDcZJQdq_TnuIYJh7g>FVt9?i^91_gDC_}opv<s<GcyGN@IxCY
zu&&J%<#oz@#n0quY(AeqzsZc;HNvbsR?4<SClUoB<;0tANJdwdlM^+fA~UGpR3(X}
zhSb7cBdkhnxs!`1chDWmZdjL0iOGg+uB3r7e-GF7@um@1+F3Q9NHAAvwyHL>^zWpI
zDx-H21|?hdPmwI8Q`u$Z)=ZJAYk{EdRXy&Dntsj~3|t5X{L#4++gdemudh$j9ig+D
zH#pesbY^eR@6%P0+krk=qYikJ2~BqgGzXs^XJt9gHa_u)Rj_XUA%|J`!&!Fb4^bTh
As{jB1

delta 680
zcmXZaF=$gk9LMqhzs4rMrfHr+5H(FoZK_}-5wVCM=uinwv4dvxQPV|B>)@b+`VvG?
zv4xW1P$VIOSlgvi;US{Yp_7WJh|s}QlnxHXMJSSj-xv=*$H5&O_xoL;;4c*Xk+Z2>
zW;lL$e@Yl(k)LbN^F56xkw0PWE!lQVo$G1aOWXQs+P$G7(M_#$PV+ottxIGMk8urW
zFaX6$NaQ1)qKccCzyWN-dMEec1?Dh=90srpJMldzQb9Q=t`u3E!XWmb9|`nfqeJ93
z?qe8P9LKsPGK&#(V|k0n9VD?2QG5=F%;PFf;sge95DA3Q@Vg>k{36d#!X!p;3>~mg
zZx{KB8eZZ&&fzF5EN^BHW#o{>5VqrokEC!HmytpOeRxOeZ-Yz~l#%hKs`@}p`JwLh
z(vS3gz<hHxR_a;Y;W<gIwwY{J<7W1j*6q=;*6W)|G%M~k_JYRT+cb0q7Z67XHCm})
z8V<H%i58|%go8uaAn*p>;2y@1#}PCOvGAJks~AP}e-Zz_Olw*U`7)OIby2O-!iRS6
Z)r{`;nH5!^sa?`>v)ZO%Ghfrwe*s1Iba?;(

diff --git a/egs/librispeech/ASR/incremental_transf/conformer.py b/egs/librispeech/ASR/incremental_transf/conformer.py
index 04f3b78ed..ca1137f31 100644
--- a/egs/librispeech/ASR/incremental_transf/conformer.py
+++ b/egs/librispeech/ASR/incremental_transf/conformer.py
@@ -510,67 +510,11 @@ class Tempformer(EncoderInterface):
     def forward(
         self, x
     ):
-        """
-        Args:
-          x:
-            The input tensor. Its shape is (batch_size, seq_len, feature_dim).
-          x_lens:
-            A tensor of shape (batch_size,) containing the number of frames in
-            `x` before padding.
-        Returns:
-          Return a tuple containing 2 tensors:
-            - embeddings: its shape is (batch_size, output_seq_len, d_model)
-            - lengths, a tensor of shape (batch_size,) containing the number
-              of frames in `embeddings` before padding.
-        """
-        x = self.encoder_embed(x)
-        x, pos_emb = self.encoder_pos(x)
-        x = x.permute(1, 0, 2)  # (N, T, C) -> (T, N, C)
-
-        lengths = (((x_lens - 1) >> 1) - 1) >> 1
-        assert x.size(0) == lengths.max().item()
-        src_key_padding_mask = make_pad_mask(lengths)
-
-        if self.dynamic_chunk_training:
-            assert (
-                self.causal
-            ), "Causal convolution is required for streaming conformer."
-            max_len = x.size(0)
-            chunk_size = torch.randint(1, max_len, (1,)).item()
-            if chunk_size > (max_len * self.short_chunk_threshold):
-                chunk_size = max_len
-            else:
-                chunk_size = chunk_size % self.short_chunk_size + 1
-
-            mask = ~subsequent_chunk_mask(
-                size=x.size(0),
-                chunk_size=chunk_size,
-                num_left_chunks=self.num_left_chunks,
-                device=x.device,
-            )
-            x = self.encoder(
-                x,
-                pos_emb,
-                mask=mask,
-                src_key_padding_mask=src_key_padding_mask,
-                warmup=warmup,
-            )  # (T, N, C)
-        else:
-            x, layer_outputs = self.encoder(
-                x,
-                pos_emb,
-                src_key_padding_mask=src_key_padding_mask,
-                warmup=warmup,
-            )  # (T, N, C)
-
-        x = x.permute(1, 0, 2)  # (T, N, C) ->(N, T, C)
-        layer_outputs = [x.permute(1, 0, 2) for x in layer_outputs]
-
-        if get_layer_output:
-            return x, lengths, layer_outputs
-        else:
-            return x, lengths
+        layer_outputs = []
+        for enum, encoder in enumerate(self.encoder_layers):
+            layer_outputs.append(encoder(x[enum]))
 
+        return layer_outputs
 
 
 class ConformerEncoderLayer(nn.Module):