training ir torchao migration #1006
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/1006
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 567ee57 with merge base 0ffbf85 ( 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: D63859678 |
Summary: Migrate capture_pre_autograd_graph to export_for_training. Differential Revision: D63859678
yushangdi
force-pushed
the
export-D63859678
branch
from
8397fff
to
850d090
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63859678 |
yushangdi
force-pushed
the
export-D63859678
branch
from
850d090
to
67fa3b4
Compare
Summary: Migrate capture_pre_autograd_graph to export_for_training. We still need to keep capture_pre_autograd_graph call because torch/ao's CI tests uses earlier version of pytorch that does not have export_for_training. Differential Revision: D63859678
|
This pull request was exported from Phabricator. Differential Revision: D63859678 |
test/dtypes/test_uint4.py
Outdated
| @@ -26,6 +25,12 @@ | |||
| ) | |||
| import copy | |||
|
|
|||
| has_export_for_training = False | |||
| try: | |||
There was a problem hiding this comment.
can this be done through a version guard? when does export_for_training start to work?
There was a problem hiding this comment.
can this be done through a version guard? when does
export_for_trainingstart to work?
it starts to exist in 2.5.0-rc1. I can make it a version guard. Thanks for the comment!
There was a problem hiding this comment.
we have some utils here btw:
Lines 25 to 29 in 0ffbf85
but notice that it's does not exactly map to current meaning of version, e.g.
TORCH_VERSION_AT_LEAST_2_5 == 2.5.0 or 2.5.0.dev
(instead of 2.5.0 + 2.6.0.dev)
Summary: Migrate capture_pre_autograd_graph to export_for_training. We still need to keep capture_pre_autograd_graph call because torch/ao's CI tests uses earlier version of pytorch that does not have export_for_training. See https://github.com/pytorch/ao/blob/main/.github/workflows/regression_test.yml Differential Revision: D63859678
yushangdi
force-pushed
the
export-D63859678
branch
from
67fa3b4
to
b5f2869
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63859678 |
Summary: Migrate capture_pre_autograd_graph to export_for_training. We still need to keep capture_pre_autograd_graph call because torch/ao's CI tests uses earlier version of pytorch that does not have export_for_training. See https://github.com/pytorch/ao/blob/main/.github/workflows/regression_test.yml Differential Revision: D63859678
yushangdi
force-pushed
the
export-D63859678
branch
from
b5f2869
to
ed15f3a
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63859678 |
Summary: Migrate capture_pre_autograd_graph to export_for_training. We still need to keep capture_pre_autograd_graph call because torch/ao's CI tests uses earlier version of pytorch that does not have export_for_training. See https://github.com/pytorch/ao/blob/main/.github/workflows/regression_test.yml Differential Revision: D63859678
yushangdi
force-pushed
the
export-D63859678
branch
from
ed15f3a
to
567ee57
Compare
|
This pull request was exported from Phabricator. Differential Revision: D63859678 |
Differential Revision: D63859678 Pull Request resolved: pytorch#1006
Summary: Migrate capture_pre_autograd_graph to export_for_training.
Differential Revision: D63859678