Fix set_acc_grad and backward wrong value

[wyg1997]

修复 set_acc_grad 后，再 backward 梯度没有正确累加的问题。

#  import oneflow as flow
import torch as flow

value = flow.tensor([[-0.0875, -0.4890,  0.9031],
            [ 0.4930, -0.6041, -1.5392]]).requires_grad_()

value.grad = flow.randn((2, 3))

output = flow.sum(value)
output.backward()
print(value.grad)

后面考虑把 AccumulateNode 改成动态绑定的形式，相关的接口就会简单很多。

[github-actions[bot]]

Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally.

[github-actions[bot]]

Speed stats:

[github-actions[bot]]

CI failed when running job: cpu-module. PR label automerge has been removed

[github-actions[bot]]

Speed stats:

[github-actions[bot]]

CI failed when running job: cpu-module. PR label automerge has been removed

[github-actions[bot]]

CI failed when running job: cpu-module. PR label automerge has been removed

[github-actions[bot]]

CI failed when running job: cuda-module. PR label automerge has been removed

[github-actions[bot]]

Speed stats:

[github-actions[bot]]

Speed stats:

[github-actions[bot]]

CI failed when running job: cuda-module. PR label automerge has been removed

[wyg1997]

python/oneflow/test/modules/test_global_nms.py:56

看起来是 nms 算法中小概率下的数据精度问题，本地 4 卡 8 卡均未复现，重跑一下。

[github-actions[bot]]

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8575/

[github-actions[bot]]

CI failed when running job: cuda-speed-test. PR label automerge has been removed