Tensor-parallel communication overlap with userbuffer backend #6362
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
erhoo82
force-pushed
the
slym/tp_overlap
branch
2 times, most recently
from
b81c170
to
609d0b1
Compare
for more information, see https://pre-commit.ci
ericharper
reviewed
Apr 11, 2023
Comment on lines
+171
to
+174
| ub_tp_comm_overlap_cfg: null | ||
| # A yaml file with userbuffer communicator configurations. This file should provide `method`, `dtype`, `num_sm`, `num_splits`, | ||
| # `cga_size`, `num_splits`, `set_sm_margin`, and `aggregate` for the communicators to use custom settings. | ||
| # If the configuration file is not provided a default setting is used for all communicators. |
There was a problem hiding this comment.
Why is this a yaml file? Should it just be
ub_tp_comm_overlap_cfg:
method: blah
dtype: blah
...
?
There was a problem hiding this comment.
The two choices were either a config object or config file since there are two many UB related args
ericharper
reviewed
Apr 11, 2023
Comment on lines
+518
to
+520
| logging.info( | ||
| "Userbuffer tensor-parallel communication overlap is available with both Transformer Engine and sequence-parallelism." | ||
| ) |
There was a problem hiding this comment.
this should be .warning and maybe add ".. only available with ... Setting ub_tp_comm_overlap to True"
ericharper
reviewed
Apr 11, 2023
Comment on lines
+333
to
+357
| # Initialize userbuffer communicators. Initialization is done only once at the | ||
| # beginning of the first training step. | ||
| if self.initialize_ub: | ||
| input_shape = [ | ||
| self.cfg.get('encoder_seq_length') * self.cfg.get('micro_batch_size'), | ||
| self.cfg.get('hidden_size'), | ||
| ] | ||
| ub_cfg_file_name = self.cfg.get('ub_tp_comm_overlap_cfg', None) | ||
| if ub_cfg_file_name is not None: | ||
| try: | ||
| import yaml | ||
|
|
||
| with open(ub_cfg_file_name, 'r') as ub_cfg_file: | ||
| ub_cfgs = yaml.safe_load(ub_cfg_file) | ||
| except (ImportError, TypeError): | ||
| print("Fail to read ub_tp_comm_overlap config file.") | ||
| else: | ||
| ub_cfgs = None | ||
| te_module.initialize_ub( | ||
| shape=input_shape, | ||
| tp_size=self.cfg.get('tensor_model_parallel_size'), | ||
| use_fp8=self.cfg.get('fp8'), | ||
| ub_cfgs=ub_cfgs, | ||
| ) | ||
| self.initialize_ub = False |
There was a problem hiding this comment.
Can this go in the .setup method then? (since it is only called once at beginning of training)
There was a problem hiding this comment.
Also, can we make it a private method and then call the it?
There was a problem hiding this comment.
To add; I'm not a fan of how we're importing the whole file from TE (which isn't a part of the API)
ericharper
requested changes
Apr 11, 2023
Signed-off-by: Sangkug Lym <slym@nvidia.com>
erhoo82
changed the title
github-actions bot
pushed a commit
that referenced
this pull request
Apr 18, 2023
* add interfaces for tp_communication overlap * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Interface to provide custom userbuffer communicator settings by yaml file * [pre-commit.ci] auto fixes from pre-commit.com hooks for more information, see https://pre-commit.ci * Jenkinsfile Signed-off-by: Sangkug Lym <slym@nvidia.com> --------- Signed-off-by: Sangkug Lym <slym@nvidia.com> Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com> Co-authored-by: Eric Harper <complex451@gmail.com>
7 tasks
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.
What does this PR do ?
Add (1) interfaces to TE and initialized (2) process group setting to support tensor-parallel communication overlap with userbuffer backend.
Changelog
Usage
Set
ub_tp_comm_overlaptoTrueBefore your PR is "Ready for review"
Pre checks:
PR Type:
If you haven't finished some of the above items you can still open "Draft" PR.
Who can review?
Anyone in the community is free to review the PR once the checks have passed.
Contributor guidelines contains specific people who can review PRs to various areas.
Additional Information