From ae4be2ce1e7fd9c8a98058b70f7fc14152ff9fb0 Mon Sep 17 00:00:00 2001 From: dohe0342 Date: Wed, 26 Apr 2023 12:29:31 +0900 Subject: [PATCH] from local --- .../.data2vec_audio.py.swp | Bin 40960 -> 40960 bytes .../data2vec_audio.py | 28 +++++++++--------- 2 files changed, 14 insertions(+), 14 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 fd253046e413f577fd6ab61fbabbf6ae648aba31..85582338e47c687b460c5480030aece708974957 100644 GIT binary patch delta 223 zcmZoTz|?SnNi4}A%+puFQqO<^2m}}y5@va%C|uYm_RgO54+{f>#$-;1=ccTj3=9W2 z7#NlUaV`*>0I;d2U^kzqlDs zR(6-1yg`{|vW>eBnAY{+nmox}8O+}5E(4V_nat{;0hLz;(?K4hlM}L8Cuex*Pd;zQ oI(deN@nlU8ZWdbw1BS^5-PI=l@KEOj*_4^4U}&{j%=1+k0A<=e2><{9 delta 199 zcmZoTz|?SnNi4}A%+puFQqO<^2m}}ybZ2>_%sjtQ?43QUJ}U!5^khzl=cd;<7#Q{g zaW@c$1F;hjO9Js;b_Rx}KwJXE8bB-!#BbOb81jHPdb6q{GwWpYSYu%YB#>H>S)!rI zFj>c~Yw|8P{mF0LL?>VH5SzT$U2d|KyTs%$ciqYD?lMp@x5?}ts*?>oL??TC7)-A9 YFqyp7Lwxdbcd^OOJk&SKdp-;U0A5oz7ytkO 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 b04515ccc..646335cb9 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 @@ -490,20 +490,20 @@ class Data2VecAudioModel(BaseFairseqModel): ## for prompt tuning if prompt is not None: - conv_feat_all = torch.tensor([]).to(features.device) - length = 0 - for i in range(padding_mask.size()[0]): - nonzero = padding_mask[i].nonzero() - try: - length += nonzero[0] - conv_feat_all = torch.cat([conv_feat_all, features[i, :nonzero[0], :]]) - except: - length += features.size()[1] - conv_feat_all = torch.cat([conv_feat_all, features[i]]) - - randint = np.random.randint(10000) - np.save(f'/home/work/workspace/icefall/egs/librispeech/ASR/conv_feat/{randint}.npy', conv_feat_all.cpu().numpy()) - exit() + if 1: + conv_feat_all = torch.tensor([]).to(features.device) + length = 0 + for i in range(padding_mask.size()[0]): + nonzero = padding_mask[i].nonzero() + try: + length += nonzero[0] + conv_feat_all = torch.cat([conv_feat_all, features[i, :nonzero[0], :]]) + except: + length += features.size()[1] + conv_feat_all = torch.cat([conv_feat_all, features[i]]) + + randint = np.random.randint(10000) + np.save(f'/home/work/workspace/icefall/egs/librispeech/ASR/conv_feat/{randint}.npy', conv_feat_all.cpu().numpy()) prompt = prompt.expand((features.size()[0], prompt.size()[0], prompt.size()[1])) features = torch.cat([prompt, features], dim=1)