Dynamo+LTC: merging related code from staging branch to master #75046
Conversation
[ghstack-poisoned]
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 1397e6a (more details on the Dr. CI page): 💚 💚 Looks good so far! There are no failures yet. 💚 💚 This comment was automatically generated by Dr. CI (expand for details).Please report bugs/suggestions to the (internal) Dr. CI Users group. |
|
Previous comments is in #75044 |
wconstab
left a comment
wconstab
left a comment
There was a problem hiding this comment.
LGTM pending some fixes commented in the previous PR. Also pending if @Krovatkin wants to review.
|
@shunting314 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
torch/csrc/lazy/python/init.cpp
Outdated
| namespace torch { | ||
| namespace lazy { | ||
|
|
||
| struct NoGilSection { |
There was a problem hiding this comment.
we should be using the pybind11 equivalent : py::gil_scoped_release
| @@ -0,0 +1,15 @@ | |||
| import torch._C._lazy | |||
There was a problem hiding this comment.
it seems a bit arbitrary that these are grouped together under computation. For example, why didn't sync_multi make the cut into this file?
There was a problem hiding this comment.
yes, I don't find a hard rule to decide which APIs goes to which module. This file specifically contains APIs needed to reused a compiled graph so far. sync_multi is a more general API thus I didn't add that here. But it should also look good to put sync_multi here
…ster" Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Differential Revision: [D35300646](https://our.internmc.facebook.com/intern/diff/D35300646) [ghstack-poisoned]
| The text can be processed by tools like dot to be rendered in pdf,png etc.""" | ||
| return torch._C._lazy._get_tensors_dot(tensors) | ||
|
|
||
| def dump_ir(tensors, ir_format): |
There was a problem hiding this comment.
@wconstab I've made the requested changes in this pr.
|
@shunting314 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
…ster" Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Differential Revision: [D35300646](https://our.internmc.facebook.com/intern/diff/D35300646) [ghstack-poisoned]
|
@shunting314 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
…ster" Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Differential Revision: [D35300646](https://our.internmc.facebook.com/intern/diff/D35300646) [ghstack-poisoned]
|
@shunting314 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
…ster" Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Differential Revision: [D35300646](https://our.internmc.facebook.com/intern/diff/D35300646) [ghstack-poisoned]
…ster" Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Differential Revision: [D35300646](https://our.internmc.facebook.com/intern/diff/D35300646) [ghstack-poisoned]
|
@shunting314 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
Summary: Pull Request resolved: #75046 Merge the code needed for dynamic+ltc integration from the staging branch to the master branch. Test Plan: Unit test ``` pytest test_extract_compiled_graph ``` test thru dynamo ``` LTC_TS_CUDA=1 time python torchbench.py --speedup-ltc -dcuda --nvfuser --randomize-input --only <model name> ``` Reviewed By: alanwaketan Differential Revision: D35300646 Pulled By: shunting314 fbshipit-source-id: 09ed20d3bb8ef80e4b93ba87ea3356a07d2dccdb
|
Hey @shunting314. |
Stack from ghstack (oldest at bottom):
Merge the code needed for dynamic+ltc integration from the staging branch to the master branch.
Test plan:
Unit test
test thru dynamo
release notes: lazy
topics: not user facing
Differential Revision: D35300646