deepspeedai / deepspeedai/DeepSpeed
[BUG] Zero3 backward_hooks use the wrong interface
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 43.1k
- Forks
- 5k
- Avg merge
- 4d 15h
- Merged PRs (30d)
- 112
Description
Describe the bug
in file ： deepspeed/runtime/zero/parameter_offload.py
` #Pre forward hook
self.forward_hooks.append(module.register_forward_pre_hook(_pre_forward_module_hook))
# Post forward hook
self.forward_hooks.append(module.register_forward_hook(_post_forward_module_hook))
# Pre backward hook
self.backward_hooks.append(module.register_forward_hook(_pre_backward_module_hook))
# post backward hook
self.backward_hooks.append(module.register_forward_pre_hook(_post_backward_module_hook))`
why is backward_hooks using interface "module.register_forward_pre_hook" and "module.register_forward_hook" ?
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start in deepspeed/runtime/zero/parameter_offload.py and inspect the four hook registrations quoted in the issue. Compare the backward-hook callbacks with the relevant PyTorch hook interfaces and trace how _pre_backward_module_hook and _post_backward_module_hook are used. Done means the interface choice is explained or corrected and Zero3 hook behavior is verified.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python, pytorch
- Domain
- distributed-systems, machine-learning
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 32/100