[CLI] Using a non-full backward hook leads to user warning.

Description Using wandb with the nn.Sequential module raises a user warning.

Wandb features wandb.watch and wandb.init.

How to reproduce Run this:

import torch
import torch.nn as nn
import wandb

with wandb.init(project="Project"):
    model = nn.Sequential(nn.Linear(100, 50),
                          nn.ReLU(),
                          nn.Linear(50, 5))
    wandb.watch(model)
    model.train()
    X = torch.randn(5, 100) # input
    criterion = nn.CrossEntropyLoss(reduction="mean")
    labels = torch.Tensor([0,1,2,3,4]).long()
    output = model(X)
    loss = criterion(output, labels)
    loss.backward()

Warning:

C:\Users\Me\anaconda3\envs\Project\lib\site-packages\torch\nn\modules\module.py:795: UserWarning: Using a non-full backward hook when the forward contains multiple autograd Nodes is deprecated and will be removed in future versions. This hook will be missing some grad_input. Please use register_full_backward_hook to get the documented behavior.
  warnings.warn("Using a non-full backward hook when the forward contains multiple autograd Nodes "

Environment

• OS: Windows 10
• Python Version: 3.8.8
• PyTorch Version: 1.8.0

[vanpelt]

The fix actually got pushed to the next release. If you want to try the fix before we release next week, you can install wandb with:

pip install --upgrade git+git://github.com/wandb/client.git@bug/issue-1122-2#egg=wandb

[github-actions[bot]]

This issue is stale because it has been open 60 days with no activity.