Break down parallelize_llama for inference cases #402
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
[ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Jun 14, 2024
ghstack-source-id: d8a32ad293ce8f1fafa141e3bbfa06654db75910 Pull Request resolved: #402
facebook-github-bot
added
the
CLA Signed
tianyu-l
requested changes
Jun 14, 2024
There was a problem hiding this comment.
Thanks for the change -- making them more modular is great!!
I had some comments -- basically I think we should make each sub-function call more modular and only pass in related arguments and configs, leaving experimental configs and interacting flags in parallelize_llama.
kwen2501
changed the title
Breaking up `parallelize_llama` into: - `apply_tp` - `apply_ac` - `apply_compile` - `apply_dp` This is for functionality reuse in inference cases, because one would not need activation checkpointing or DP there. Can also improve code modularity and readability. [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Jun 14, 2024
ghstack-source-id: 9aeee4c063c63eed380cac219c9c8e1eb4169f9d Pull Request resolved: #402
tianyu-l
reviewed
Jun 14, 2024
Breaking up `parallelize_llama` into: - `apply_tp` - `apply_ac` - `apply_compile` - `apply_dp` This is for functionality reuse in inference cases, because one would not need activation checkpointing or DP there. Can also improve code modularity and readability. [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Jun 14, 2024
ghstack-source-id: 72e37e2e506af6115f9cb18179543dd6df602961 Pull Request resolved: #402
tianyu-l
approved these changes
Jun 14, 2024
tianyu-l
reviewed
Jun 14, 2024
Comment on lines
303
to
305
| if job_config.model.norm_type == "fused_rmsnorm": | ||
| raise NotImplementedError( | ||
| "fused_rmsnorm not yet compatible with TP. Please use layernorm or rmsnorm." |
There was a problem hiding this comment.
this can be removed thanks to #404
There was a problem hiding this comment.
Yeah, noticed in CI. Removed.
Breaking up `parallelize_llama` into: - `apply_tp` - `apply_ac` - `apply_compile` - `apply_dp` This is for functionality reuse in inference cases, because one would not need activation checkpointing or DP there. Can also improve code modularity and readability. [ghstack-poisoned]
kwen2501
added a commit
that referenced
this pull request
Jun 14, 2024
ghstack-source-id: fc8e221b5047337f59dea31f2c51d6168fe4fe88 Pull Request resolved: #402
kwen2501
added a commit
that referenced
this pull request
Jun 14, 2024
ghstack-source-id: fc8e221b5047337f59dea31f2c51d6168fe4fe88 Pull Request resolved: #402
tianyu-l
pushed a commit
to tianyu-l/torchtitan_intern24
that referenced
this pull request
Aug 16, 2024
ghstack-source-id: fc8e221b5047337f59dea31f2c51d6168fe4fe88 Pull Request resolved: pytorch#402
philippguevorguian
pushed a commit
to YerevaNN/YNNtitan
that referenced
this pull request
Aug 17, 2024
ghstack-source-id: fc8e221b5047337f59dea31f2c51d6168fe4fe88 Pull Request resolved: pytorch#402
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.
Stack from ghstack (oldest at bottom):
Breaking up
parallelize_llamainto:apply_tpapply_acapply_compileapply_dpThis is for functionality reuse in inference cases, because one would not need activation checkpointing or DP there.
Can also improve code modularity and readability.