DSTK / semantic_tokenizer /f40ms /fairseq_npu_patch.py
gooorillax's picture
first push of codes and models for g2p, t2u, tokenizer and detokenizer
cd8454d
# Copyright (C) 2025. Huawei Technologies Co., Ltd. All Rights Reserved. (authors: Xiao Chen)
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
# http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from patch_utils import MindSpeedPatchesManager as aspm
import torch_npu
def cnn_feature_forward_npu(self, x):
# BxT -> BxCxT
x = x.unsqueeze(1)
for conv in self.conv_layers:
# x = conv(x)
x_tmp = conv[:-1](x)
x = torch_npu.npu_gelu(x_tmp)
return x
def patch_for_npu():
# replace torch.cuda.get_device_capability with implementation from MindSpeed
aspm.register_patch('fairseq.models.wav2vec.wav2vec2.ConvFeatureExtractionModel.forward', cnn_feature_forward_npu)
aspm.apply_patches()