Enable non-ZeRO mode #7515
Merged
Enable non-ZeRO mode #7515
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Collaborator
Author
|
@tohtana I have extended your |
Collaborator
Author
|
@stas00 FYI. |
Signed-off-by: Olatunji Ruwase <[email protected]>
Collaborator
|
@sfc-gh-truwase Can you check why the cpu-torch unit tests are failing? Also can we say for non-ZeRO training, we fully rely on |
Collaborator
|
The elif model_dtype == grad_accum_dtype:
if model_dtype == torch.bfloat16:
if self.pipeline_parallelism:
logger.warning(...)
return BFLOAT16
else:
raise NotImplementedError(...)It only allows BF16 accumulation when PP is enabled. |
Collaborator
Author
Collaborator
Author
@PKUWZP, great question. No, non-ZeRO can also use native mixed-precision training. Below is how this PR expands the mixed precision training options of DeepSpeed: 
|
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
stas00
approved these changes
Aug 27, 2025
Collaborator
stas00
left a comment
stas00
left a comment
There was a problem hiding this comment.
Thank you for making stage=0 work, Tunji
The PR is looking good, added a few minor suggestions
Co-authored-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Co-authored-by: Stas Bekman <[email protected]>
Signed-off-by: Olatunji Ruwase <[email protected]>
delock
pushed a commit
that referenced
this pull request
Sep 1, 2025
Enabled via `stage=0` which corresponds to DDP. Remove hardwired path to b16_optimizer. Enable`torch.autocast` for DDP training Enable native mixed precision DDP for bfloat16 Update torch.autocast and native mixed precision UTs <img width="976" height="184" alt="image" src="https://github.com/user-attachments/assets/92904cdc-e312-46a4-943f-011eb5ab146a" /> --------- Signed-off-by: Olatunji Ruwase <[email protected]> Co-authored-by: Stas Bekman <[email protected]> Signed-off-by: Ma, Guokai <[email protected]>
mauryaavinash95
pushed a commit
to DataStates/DeepSpeed
that referenced
this pull request
Oct 4, 2025
Enabled via `stage=0` which corresponds to DDP. Remove hardwired path to b16_optimizer. Enable`torch.autocast` for DDP training Enable native mixed precision DDP for bfloat16 Update torch.autocast and native mixed precision UTs <img width="976" height="184" alt="image" src="https://github.com/user-attachments/assets/92904cdc-e312-46a4-943f-011eb5ab146a" /> --------- Signed-off-by: Olatunji Ruwase <[email protected]> Co-authored-by: Stas Bekman <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Enabled via
stage=0which corresponds to DDP.Remove hardwired path to b16_optimizer.
Enable
torch.autocastfor DDP trainingEnable native mixed precision DDP for bfloat16
Update torch.autocast and native mixed precision UTs