New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
CPU SHM based inference_all_reduce improve #5320
Merged
Merged
Conversation
This file contains 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
tjruwase
approved these changes
Apr 1, 2024
rraminen
pushed a commit
to ROCm/DeepSpeed
that referenced
this pull request
May 9, 2024
This PR improves SHM based inference_all_reduce on CPU: 1. Optimize for larger message size which affects performance of first token generation with long context. For example, for llama2 70b with 1024 input sequence length, all_reduce message size is 32MB with single batch. * Increased SHM buffer size from 1MB/worker to 32MB/worker * Each worker allocate SHM buffer on its own NUMA node, instead of rank 0 allcoate SHM buffers for all other workers * For message size > 1MB, a more distributed algoritm is used to make memory bandwidth and computation evenly distributed among workers 2. Decouple SHM based collective code with oneCCL based code, making it ready to integrate with other backend i.e. gloo backend 3. Loosen the condition SHM based allreduce is used, i.e. message size does not have to divisible by 32 bytes. The new distributed algorithm, combine with larger per worker SHM buffer, brings ~3x allreduce performance improvement for 32MB message size on a 2 socket machine. --------- Co-authored-by: Logan Adams <114770087+loadams@users.noreply.github.com> Co-authored-by: Logan Adams <loadams@microsoft.com> Co-authored-by: Olatunji Ruwase <olruwase@microsoft.com>
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 improves SHM based inference_all_reduce on CPU:
The new distributed algorithm, combine with larger per worker SHM buffer, brings ~3x allreduce performance improvement for 32MB message size on a 2 socket machine.