enable to_mxfp8 cast for DTensor #2420
Merged
+168
−18
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
|
Stack from ghstack (oldest at bottom): |
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/2420
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit aabeb61 with merge base b963540 ( This comment was automatically generated by Dr. CI and updates every 15 minutes. |
facebook-github-bot
added
the
CLA Signed
vkuzo
added a commit
that referenced
this pull request
Jun 21, 2025
Summary: 1. add a test for casting a DTensor to mxfp8 2. make the test pass: a. remove addition of epsilon, it's not supported in DTensor world but we also don't need it anymore since we are no longer using `log2` anywhere. b. replace `<<` with `torch.bitwise_left_shift` and `>>` with `torch.bitwise_right_shift`. The short versions are silently broken for DTensor inputs, but the verbose versions work. 3. set up the wiring for testing mxfp8 with TP on a toy model. Note that making this work is split for the next PR, as this PR got too large. Test Plan: ```bash ./test/prototype/mx_formats/test_dtensor.sh ./test/float8/test_dtensor.sh pytest test/prototype/mx_formats/ ``` Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: 756da75 ghstack-comment-id: 2993264092 Pull Request resolved: #2420
vkuzo
added a commit
that referenced
this pull request
Jun 21, 2025
Summary: 1. add a test for casting a DTensor to mxfp8 2. make the test pass: a. remove addition of epsilon, it's not supported in DTensor world but we also don't need it anymore since we are no longer using `log2` anywhere. b. replace `<<` with `torch.bitwise_left_shift` and `>>` with `torch.bitwise_right_shift`. The short versions are silently broken for DTensor inputs, but the verbose versions work. 3. set up the wiring for testing mxfp8 with TP on a toy model. Note that making this work is split for the next PR, as this PR got too large. Test Plan: ```bash ./test/prototype/mx_formats/test_dtensor.sh ./test/float8/test_dtensor.sh pytest test/prototype/mx_formats/ ``` Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: e054ed4 ghstack-comment-id: 2993264092 Pull Request resolved: #2420
vkuzo
added a commit
that referenced
this pull request
Jun 21, 2025
Summary: 1. add a test for casting a DTensor to mxfp8 2. make the test pass: a. remove addition of epsilon, it's not supported in DTensor world but we also don't need it anymore since we are no longer using `log2` anywhere. b. replace `<<` with `torch.bitwise_left_shift` and `>>` with `torch.bitwise_right_shift`. The short versions are silently broken for DTensor inputs, but the verbose versions work. 3. set up the wiring for testing mxfp8 with TP on a toy model. Note that making this work is split for the next PR, as this PR got too large. Test Plan: ```bash ./test/prototype/mx_formats/test_dtensor.sh ./test/float8/test_dtensor.sh pytest test/prototype/mx_formats/ ``` Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: d82a83f ghstack-comment-id: 2993264092 Pull Request resolved: #2420
vkuzo
added a commit
that referenced
this pull request
Jun 21, 2025
Summary: 1. add a test for casting a DTensor to mxfp8 2. make the test pass: a. remove addition of epsilon, it's not supported in DTensor world but we also don't need it anymore since we are no longer using `log2` anywhere. b. replace `<<` with `torch.bitwise_left_shift` and `>>` with `torch.bitwise_right_shift`. The short versions are silently broken for DTensor inputs, but the verbose versions work. 3. set up the wiring for testing mxfp8 with TP on a toy model. Note that making this work is split for the next PR, as this PR got too large. Test Plan: ```bash ./test/prototype/mx_formats/test_dtensor.sh ./test/float8/test_dtensor.sh pytest test/prototype/mx_formats/ ``` Reviewers: Subscribers: Tasks: Tags: ghstack-source-id: cd43d13 ghstack-comment-id: 2993264092 Pull Request resolved: #2420
vkuzo
added
the
topic: improvement
danielvegamyhre
approved these changes
Jun 24, 2025
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.
Summary:
a. remove addition of epsilon, it's not supported in DTensor world but
we also don't need it anymore since we are no longer using
log2anywhere.
b. replace
<<withtorch.bitwise_left_shiftand>>withtorch.bitwise_right_shift. The short versions are silently brokenfor DTensor inputs, but the verbose versions work. See
<<and>>operators seem silently broken for DTensor operand 1 and scalar operand 2 pytorch#156533.making this work is split for the next PR, as this PR got too large.
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags: