RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling `cublasCreate(handle)`

pycharm连接Autodl服务器，运行报错

RuntimeError: CUDA error: CUBLAS_STATUS_NOT_INITIALIZED when calling `cublasCreate(handle)`

很多帖子说batch size太大，或者CUDA版本和torch不匹配，尝试无果。

    def forward(self, x):
        #print(x.shape)
        x = self.Block1(x)
        #print(x.shape)
        x = self.Block2(x)
        #print(x.shape)
        x = self.Block3(x)
        #print(x.shape)
        x = x.view(x.size(0), -1)
        #print(x.shape)
        x = self.fc(x)
        x = nn.LayerNorm(x.size())(x.cpu())
        x = x.cuda()
        #print(x.shape)
        return x

最后把网络的各个shape打印出来，发现其实是数据shape不匹配。

 self.fc = nn.Sequential(
            nn.Linear(5760, 18)
 )

出现这个问题先去看数据、网络维度