Enable ZeRO-3 linear wrapper for existing models - #8189
Conversation
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 455574c290
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
| else: | ||
| grad_bias = grad_output.sum(0) | ||
| return grad_input, grad_weight, grad_bias | ||
| weight_was_partitioned = (hasattr(weight, "ds_status") |
There was a problem hiding this comment.
It occurs to me that this better be a function of weight as well i.e. weight.is_partitioned(), maybe worth a seperate PR.
There was a problem hiding this comment.
How about using is_zero_param() for detection?
|
Hi @tohtana I have left my comments. How many memory we may save from this PR? Is a documentation change needed? |
3 participants
Passing an already-constructed model to
deepspeed.initialize()with ZeRO-3 andmemory_efficient_linear=truedoes not install the ZeRO-3 Linear wrapper. The wrapper is currently installed only when the model is constructed inside adeepspeed.zero.Init()context.Without the wrapper, the standard Linear implementation can retain the gathered weight storage until backward completes, significantly increasing memory usage.
This PR activates the existing ZeRO-3 Linear wrapper for the
deepspeed.initialize(model=...)path whenmemory_efficient_linear=true, without requiring adeepspeed.zero.Init()context.