Refactor dtype handling in export_llama #9430
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
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/9430
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New Failure, 2 Unrelated FailuresAs of commit 62f1e9d with merge base a828307 ( NEW FAILURE - The following job has failed:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
jackzhxng
force-pushed
the
export-D71515138
branch
from
21b73fb to
21941b6
Compare
jackzhxng
added a commit
to jackzhxng/executorch
that referenced
this pull request
Mar 20, 2025
Summary: Pull Request resolved: pytorch#9430 Differential Revision: D71515138
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
jackzhxng
changed the title
jackzhxng
added a commit
to jackzhxng/executorch
that referenced
this pull request
Mar 20, 2025
Summary: Pull Request resolved: pytorch#9430 Differential Revision: D71515138
jackzhxng
force-pushed
the
export-D71515138
branch
from
21941b6 to
fa12dfa
Compare
jackzhxng
added a commit
to jackzhxng/executorch
that referenced
this pull request
Mar 20, 2025
Summary: Pull Request resolved: pytorch#9430 Differential Revision: D71515138
jackzhxng
force-pushed
the
export-D71515138
branch
from
fa12dfa to
73809d0
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
jackzhxng
added a commit
to jackzhxng/executorch
that referenced
this pull request
Mar 20, 2025
Summary: No more converting from fp32 -> checkpoint dtype (fp16 or lower) -> back to dtype override (fp32), where we are losing precision on buffers. Also cleans up the entire dtype, now it only occurs outside of model.py, who's responsibility should just be for loading the model. Differential Revision: D71515138
jackzhxng
force-pushed
the
export-D71515138
branch
from
73809d0 to
6be43cf
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
jackzhxng
added a commit
to jackzhxng/executorch
that referenced
this pull request
Mar 20, 2025
Summary: While it might make sense intuitively to have the dtype of the model be the dtype of the checkpoint, this isn't possible for all backends which only support some dtypes. We need to be explicit about the dtype of the model for this reason. No more intermediate conversion into the checkpoint dtype, which could cause precision loss in situations like these: fp32 -> checkpoint dtype (fp16 or lower) -> back to dtype override (fp32), where we are losing precision on buffers that are instantiated in fp32 and downcast to fp16. Differential Revision: D71515138
jackzhxng
force-pushed
the
export-D71515138
branch
from
6be43cf to
e112ad8
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
Summary: While it might make sense intuitively to have the dtype of the model be the dtype of the checkpoint, this isn't possible for all backends which only support some dtypes. We need to be explicit about the dtype of the model for this reason. No more intermediate conversion into the checkpoint dtype, which could cause precision loss in situations like these: fp32 -> checkpoint dtype (fp16 or lower) -> back to dtype override (fp32), where we are losing precision on buffers that are instantiated in fp32 and downcast to fp16. Reviewed By: kimishpatel Differential Revision: D71515138
jackzhxng
force-pushed
the
export-D71515138
branch
from
e112ad8 to
62f1e9d
Compare
|
This pull request was exported from Phabricator. Differential Revision: D71515138 |
kimishpatel
approved these changes
Mar 21, 2025
DannyYuyang-quic
pushed a commit
to CodeLinaro/executorch
that referenced
this pull request
Apr 2, 2025
Differential Revision: D71515138 Pull Request resolved: pytorch#9430
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.
Differential Revision: D71515138