Fix ZeRO stage 1 and add stage 2 support with DeepCompile #7366
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: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
Signed-off-by: Masahiro Tanaka <[email protected]>
…eed into tohtana/dc_z1_no_sync
This was referenced Jun 23, 2025
loadams
approved these changes
Jun 27, 2025
tohtana
added a commit
that referenced
this pull request
Jun 27, 2025
This PR improves the coverage of DeepCompile. - Use real parameters when recompilation happens - Handling overflow error in profiling This PR should be merged after #7366. ZeRO1 and ZeRO3 both worked with OpenRLHF. See [Wiki page](https://github.com/tohtana/DeepCompile_docs/wiki/Debug-with-OpenRLHF-(%237243)) for more details. --------- Signed-off-by: Masahiro Tanaka <[email protected]> Co-authored-by: Stas Bekman <[email protected]>
lpnpcs
pushed a commit
to lpnpcs/DeepSpeed
that referenced
this pull request
Jul 30, 2025
…i#7366) This PR fixes the behavior of DeepCompile's ZeRO stage 1 and adds stage 2 support. DeepCompile's ZeRO1 currently performs allreduce at every iteration even when it is not a gradient accumulation boundary. This significantly slows down the performance when gradient accumulation is enabled. This PR fixes this issue by performing allreduce only at the gradient accumulation boundary. As the current behavior is similar to ZeRO2, this PR also adds DeepCompile's ZeRO2 support. We can now set zero stage to 2 with DeepCompile. The loss values, performance, and memory usages were verified using this [verification tool](https://github.com/tohtana/ds_verify_loss) ([results](https://github.com/tohtana/ds_verify_loss/blob/main/results/results_20250617_035117/report.md)). --------- Signed-off-by: Masahiro Tanaka <[email protected]> Co-authored-by: Stas Bekman <[email protected]> Co-authored-by: Logan Adams <[email protected]>
lpnpcs
pushed a commit
to lpnpcs/DeepSpeed
that referenced
this pull request
Jul 30, 2025
This PR improves the coverage of DeepCompile. - Use real parameters when recompilation happens - Handling overflow error in profiling This PR should be merged after deepspeedai#7366. ZeRO1 and ZeRO3 both worked with OpenRLHF. See [Wiki page](https://github.com/tohtana/DeepCompile_docs/wiki/Debug-with-OpenRLHF-(%237243)) for more details. --------- Signed-off-by: Masahiro Tanaka <[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.
This PR fixes the behavior of DeepCompile's ZeRO stage 1 and adds stage 2 support.
DeepCompile's ZeRO1 currently performs allreduce at every iteration even when it is not a gradient accumulation boundary. This significantly slows down the performance when gradient accumulation is enabled. This PR fixes this issue by performing allreduce only at the gradient accumulation boundary.
As the current behavior is similar to ZeRO2, this PR also adds DeepCompile's ZeRO2 support. We can now set zero stage to 2 with DeepCompile.
The loss values, performance, and memory usages were verified using this verification tool (results).