Centralize Async TP Enablement with maybe_enable_async_tp API #1619
+32
−51
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
This PR addresses duplicated code related to enabling async TP across different parts of the codebase. It introduces a new API, `maybe_enable_async_tp()`, which centralizes the enablement logic and is reused consistently in all models. Note that while this PR fixes one async TP bug in TorchTitan, it does not fully resolve #1613, as there appear to be additional bugs in PyTorch's async TP implementation.
meta-cla
bot
added
the
CLA Signed
tianyu-l
reviewed
Aug 21, 2025
| @@ -139,12 +133,26 @@ def parallelize_llama( | |||
| return model | |||
|
|
|||
|
|
|||
| def maybe_enable_async_tp(job_config: JobConfig, tp_mesh: DeviceMesh): | |||
There was a problem hiding this comment.
I'd suggest we take this chance to put it into model-agnostic file. Specifically I'm thinking of torchtitan/distributed/tensor_parallel.py where we can also put NoParallel (https://github.com/pytorch/torchtitan/blob/main/torchtitan/distributed/expert_parallel.py#L116) to.
I'm also thinking we may want to put most apply_ac (and maybe apply_compile) logic to that folder, as they are pretty much the same across all models.
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Outdated suggestions cannot be applied.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 addresses duplicated code related to enabling async TP across different parts of the codebase. It introduces a new API,
maybe_enable_async_tp(), which centralizes the enablement logic and is reused consistently in all models.Note that while this PR fixes one async TP bug in TorchTitan, it does not fully resolve #1613, as there appear to be additional bugs in PyTorch's async TP implementation.