迁移网络任务-tacotron2时mindspore的权重初始化与torch的不一致

经验分享 Tech Blogs 功能调试-Function Debugging
1

1.系统环境

硬件环境(Ascend/GPU/CPU): GPU
MindSpore版本: mindspore=2.0
执行模式（PyNative/ Graph）:不限
Python版本：3.7
操作系统平台：Linux

2. 问题描述

迁移网络任务-tacotron2时mindspore的权重初始化与torch的不一致

cke_1019.png
cke_1019.png554×161 59.5 KB

代码：

ts.migrator.compare_pth_and_ckpt(  
    weight_map_path="/home/hodia/python/Pytorch-fasterRCNN/pt_net_info/torch_net_map.json",  
    pt_file_path="/home/hodia/python/Pytorch-fasterRCNN/pt_net_info/torch_troubleshooter_create.pth",  
    ms_file_path="/home/hodia/python/MindSpore-fasterRCNN/ms_net_info/mindspore_troubleshooter_create.ckpt",  
compare_value=True)

3. 解决方案

将pytorch的权重转化为mindspore，进行训练

ts.migrator.convert_weight_and_load(weight_map_path="/home/hodia/python/Pytorch-fasterRCNN/pt_net_info/torch_net_map.json", 
                                    pt_file_path="/home/hodia/python/Pytorch-fasterRCNN/pt_net_info/torch_troubleshooter_create.pth", 
                                    net=model)

主要表现在初始化中
Torch的初始化

for layer in self.children():  
    if isinstance(layer,nn.Conv2d):  
        torch.nn.init.normal_(layer.weight,std=0.01)  
        torch.nn.init.constant_(layer.bias,0)

MindSpore的初始化

for _, layer in self.cells_and_names():  
    if isinstance(layer,nn.Conv2d):  
        layer.weight.set_data(ms.common.initializer.initializer(ms.common.initializer.Normal(sigma=0.01,mean=0.0),layer.weight.shape,layer.weight.dtype))  
        layer.bias.set_data(ms.common.initializer.initializer("zeros",layer.bias.shape,layer.bias.dtype))