Add adaptive gather-buffer pooling for DeepCompile ZeRO-3 - #8234
Conversation
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
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: b26fccd563
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".
| @@ -4,6 +4,7 @@ | |||
| // DeepSpeed Team | |||
|
|
|||
| #include "deepcompile.h" | |||
There was a problem hiding this comment.
This is a non-merge commit, but its commit message has no Signed-off-by line, so it does not satisfy the repository's mandatory commit/CI requirement and should be signed off before merging.
AGENTS.md reference: AGENTS.md:L8-L8
Useful? React with 👍 / 👎.
|
This looks like a solid, focused change. |
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
Signed-off-by: Masahiro Tanaka <mtanaka@anyscale.com>
2 participants
(#8233 needs to be merged before this PR)
Splits the native gather-buffer pool and allocator-pressure recovery changes out of #8169. This is stacked on the scheduler PR and passed 12/12 tests on a single node with 2 H100 GPUs.