RuntimeError: grad can be implicitly created only for scalar outputs

[Ression-LZ]

Hi, I tried to run your code, but there is a problem:
File "/home/lamo/PycharmProjects/FedSL-master/models/UpdateSplit.py", line 72, in train clientH.backward(h_grads)
RuntimeError: grad can be implicitly created only for scalar outputs
Could you please tell me how to fix it?

[abedidev]

Please consider modifying your labels to scalars. Class labels must be scalar/discrete numbers and cannot be real-valued or float values.

…

On Thu, Jan 5, 2023 at 9:43 AM Lamo ***@***.***> wrote: Hi, I tried to run your code, but there is a problem: File "/home/lamo/PycharmProjects/FedSL-master/models/UpdateSplit.py", line 72, in train clientH.backward(h_grads) RuntimeError: grad can be implicitly created only for scalar outputs Could you please tell me how to fix it? — Reply to this email directly, view it on GitHub <#1>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ARRYAPGJQLZMMXAQH7P4R5LWQ3MYXANCNFSM6AAAAAATSA57BA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[tingfengwuyou]

您好，我尝试运行您的代码，但出现问题： 文件“/home/lamo/PycharmProjects/FedSL-master/models/UpdateSplit.py”，第 72 行，在 train clientH.backward(h_grads) RuntimeError: grad can仅为标量输出隐式创建 你能告诉我如何修复它吗？

请问你解决了吗好朋友

[Ression-LZ]

您好，我尝试运行您的代码，但出现问题： 文件“/home/lamo/PycharmProjects/FedSL-master/models/UpdateSplit.py”，第 72 行，在 train clientH.backward(h_grads) RuntimeError: grad can仅为标量输出隐式创建 你能告诉我如何修复它吗？

请问你解决了吗好朋友

没有捏，你解决了吗好朋友？

[hunter123520]

Hello everyone, for people who still struggle with that error, i find out that replacing
h_grads = h.grad
with
h_grads = torch.autograd.grad(loss, h, retain_graph=True)[0]
in the UpdateSplit file in the train function of the LocalUpdate class, will fix the problem. Good Luck.