Skip to content

m-fsdp: wire use_precision_aware_optimizer from ddp_config to ParamAn… - #4024

Merged
cspades merged 4 commits into
NVIDIA:mainfrom
rapatel:ripatel/fsdp_prec_optim_flag
Apr 1, 2026
Merged

m-fsdp: wire use_precision_aware_optimizer from ddp_config to ParamAn…#4024
cspades merged 4 commits into
NVIDIA:mainfrom
rapatel:ripatel/fsdp_prec_optim_flag

Conversation

@rapatel

@rapatel rapatel commented Mar 25, 2026

Copy link
Copy Markdown
Contributor

…dGradBuffer

What does this PR do ?

⚠️ For major changes (either in lines of code or in its impact), please make sure to first share a design doc with the team. If you're unsure what's the best way to do so, contact the @mcore-oncall.

Contribution process

Pre-checks

  • I have added relevant unit tests
  • I have added relevant functional tests
  • I have added proper typing to my code Typing guidelines
  • I have added relevant documentation
  • I have run the autoformatter.sh on my PR

Code review

Feel free to message or comment the @mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!

All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.

Step 1: Mark PR as "Ready for Review"

  1. When your PR is ready, click Ready for Review.
  2. An oncall reviewer is auto-assigned and expert reviewers are notified based on your changes.
    • Some PRs may jump straight to step 2. This is determined by .github/CODEOWNERS.

⚠️ Only mark as ready once merge-conflicts are resolved and the CI is passing.
Final Review might get declined if these requirements are not fulfilled.

Step 2: Final Review

For PRs that change megatron/core, once all expert reviewers have approved, the Final Review label is applied automatically and final reviewers are assigned.

For PRs outside megatron/core, this step is skipped.

Step 3: Approved

Once all required reviewers have approved, the Approved label is applied automatically.

Merge

Any member of mcore-engineers will be able to merge your PR.

For MRs into `dev` branch The proposed review process for `dev` branch is under active discussion.

MRs are mergable after one approval by either eharper@nvidia.com or zijiey@nvidia.com.

@rapatel
rapatel requested review from a team as code owners March 25, 2026 05:28
@copy-pr-bot

copy-pr-bot Bot commented Mar 25, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@svcnvidia-nemo-ci
svcnvidia-nemo-ci marked this pull request as draft March 25, 2026 05:28
@github-actions

Copy link
Copy Markdown
Contributor

This PR has been automatically converted to draft because all PRs must start as drafts.

When you are ready for review, click Ready for Review to begin the review process. This will:

  1. Add the oncall reviewer (optional reviewer)
  2. Add required review teams based on your changes

See the contribution guide for more details.

@shjwudp
shjwudp marked this pull request as ready for review March 25, 2026 05:30
@svcnvidia-nemo-ci
svcnvidia-nemo-ci requested a review from a team March 25, 2026 05:30
@shjwudp

shjwudp commented Mar 25, 2026

Copy link
Copy Markdown
Contributor

/ok to test 74ddff5

@svcnvidia-nemo-ci svcnvidia-nemo-ci added this to the Core 0.16 milestone Mar 25, 2026
@svcnvidia-nemo-ci svcnvidia-nemo-ci added the Final Review PR is in the "final review" stage label Mar 25, 2026
denys-fridman added a commit to denys-fridman/Megatron-LM that referenced this pull request Mar 25, 2026

@cspades cspades left a comment

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nit: Needs README / PyPI package documentation for the new argument, and possibly any arguments in training.py should be set during DDPConfig init.

Can handle in a subsequent PR if not done here, thanks!

"""If true, Megatron-FSDP's ParamAndGradBuffer uses the precision-aware optimizer
gradient path (e.g. `decoupled_grad` on optimizer parameters) instead of casting
main gradients to parameter dtype for `.grad`. Should match the training optimizer
setting (e.g. `--use-precision-aware-optimizer`).

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

command line args are only one way these dataclasses get populated, we shouldn't reference them in these docs. Is there an OptimizerConfig dataclass that has this option that should match?

@rapatel rapatel Mar 25, 2026

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do see this - https://github.com/NVIDIA/Megatron-LM/blob/main/megatron/core/optimizer/optimizer_config.py#L187:

use_precision_aware_optimizer: bool = False """If true, allows optimizer-related tensors (master_param, gradients and optimizer states) to be set to lower precision. Defaults to False. """

@cspades cspades Mar 25, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Currently I think Megatron-Bridge and Megatron-FSDP only have this DDPConfig to shove all arguments into.

I think let's remove the reference to --use-precision-aware-optimizer to avoid confusing MLM vs. MBridge vs. etc. users of MCore, but I think having the same name as the attr in the OptimizerConfig might be helpful since they should be used together (when relevant, of course). Generally DDP-type classes need to know where to dump the grad.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I've gone ahead and updated the docstring.

@rapatel
rapatel force-pushed the ripatel/fsdp_prec_optim_flag branch from 74ddff5 to e9a02e3 Compare March 25, 2026 18:49
@gautham-kollu

Copy link
Copy Markdown
Contributor

/ok to test e9a02e3

Comment thread megatron/core/distributed/distributed_data_parallel_config.py Outdated
@cspades
cspades force-pushed the ripatel/fsdp_prec_optim_flag branch from e9a02e3 to 7ef00ed Compare March 31, 2026 17:47
@cspades
cspades requested review from a team as code owners March 31, 2026 17:47
@svcnvidia-nemo-ci svcnvidia-nemo-ci removed the Final Review PR is in the "final review" stage label Mar 31, 2026
@cspades
cspades force-pushed the ripatel/fsdp_prec_optim_flag branch from 7ef00ed to 9e9aaab Compare March 31, 2026 17:50
… MLM --use-precision-aware-optimizer.

Signed-off-by: Cory Ye <cye@nvidia.com>
@cspades
cspades force-pushed the ripatel/fsdp_prec_optim_flag branch from 9e9aaab to cd0bf5e Compare March 31, 2026 17:56
@cspades

cspades commented Mar 31, 2026

Copy link
Copy Markdown
Member

/ok to test 4d8cf30

@cspades

cspades commented Mar 31, 2026

Copy link
Copy Markdown
Member

/ok to test 038b141

@jaredcasper jaredcasper left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving, but can we make a plan for how to remove the megatron_fsdp_* args from ddpconfig?

@cspades

cspades commented Apr 1, 2026

Copy link
Copy Markdown
Member

Approving, but can we make a plan for how to remove the megatron_fsdp_* args from ddpconfig?

Yeah, I'll start a design doc with Megatron-Bridge and see if we can pass in DDPConfig and FSDPConfig separately, that's the main blocker at this point. Discussions can be had there.

@svcnvidia-nemo-ci svcnvidia-nemo-ci added the Approved All necessary approvals have been made label Apr 1, 2026
@cspades
cspades enabled auto-merge April 1, 2026 20:24
@cspades
cspades added this pull request to the merge queue Apr 1, 2026
@svcnvidia-nemo-ci

Copy link
Copy Markdown
Contributor

🔄 Merge queue validation started!

You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/23869171336

Merged via the queue into NVIDIA:main with commit fe5291f Apr 1, 2026
66 of 68 checks passed
yangbofun pushed a commit to xlm-research/Megatron-LM that referenced this pull request May 22, 2026
NVIDIA#4024)

Signed-off-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
yhgalaxy pushed a commit to yhgalaxy/Megatron-LM that referenced this pull request Jun 17, 2026
NVIDIA#4024)

Signed-off-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
Signed-off-by: yhgalaxy <yhgalaxy@outlook.com>
jon-barker pushed a commit to jon-barker/Megatron-LM that referenced this pull request Jul 10, 2026
NVIDIA#4024)

Signed-off-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
Signed-off-by: Jon Barker <jbarker@aws-cmh-slurm-1-vscode-02.cm.cluster>
Caozhou1995 pushed a commit to flagos-ai/Megatron-LM-FL that referenced this pull request Jul 17, 2026
# Description
NVIDIA#4024
NVIDIA#3746
NVIDIA#2799
NVIDIA#3676
NVIDIA#4133
NVIDIA#3295
NVIDIA#3831
NVIDIA#3754


## Type of change

- [ ] New feature (non-breaking change which adds functionality)
- [ ] Infra/Build change (changes to CI/CD workflows or build scripts)
- [ ] Code refactoring
- [ ] Documentation change
- [x] Bug fix
- [ ] Breaking change

## Changes

- Content 1
- Content 2
- Content 3
- Content 4

## Checklist

- [ ] I have read and followed the contributing guidelines
- [ ] The functionality is complete
- [ ] I have commented my code, particularly in coverage report
uploading steps
- [ ] I have made corresponding changes to the documentation
- [ ] My changes generate no new warnings
- [ ] I have added/updated tests that prove my feature works
- [ ] New and existing unit tests pass locally

---------

Signed-off-by: Cory Ye <cye@nvidia.com>
Signed-off-by: yexiaochuan <yexiaochuan@baidu.com>
Signed-off-by: Akshat Kumar <akshat230405@gmail.com>
Signed-off-by: Youngeun Kwon <youngeunk@nvidia.com>
Signed-off-by: jinliangl <jinliangl@nvidia.com>
Co-authored-by: Ritesh Patel <ripatel@nvidia.com>
Co-authored-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
Co-authored-by: Xiaochuan Ye <tap91624@gmail.com>
Co-authored-by: yexiaochuan <yexiaochuan@baidu.com>
Co-authored-by: Akshat8510 <akshat230405@gmail.com>
Co-authored-by: Kunlun Li <94586211+kunlunl@users.noreply.github.com>
Co-authored-by: eternally-z <zzywzj@gmail.com>
Co-authored-by: Philip Petrakian <ppetrakian@nvidia.com>
Co-authored-by: Youngeun Kwon <youngeunk@nvidia.com>
Co-authored-by: xuwchen <xuwenc@nvidia.com>
Co-authored-by: Li Jinliang <jinliangl@nvidia.com>
terminator123 pushed a commit to 021ai/Megatron-LM that referenced this pull request Aug 3, 2026
NVIDIA#4024)

Signed-off-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <cye@nvidia.com>
Co-authored-by: Cory Ye <44509866+cspades@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Approved All necessary approvals have been made complexity: low module: megatron-fsdp

Projects

None yet

Development

Successfully merging this pull request may close these issues.

9 participants