-
Notifications
You must be signed in to change notification settings - Fork 22.3k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Disable meta device tests. #74468
Disable meta device tests. #74468
Conversation
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyang@fb.com> [ghstack-poisoned]
CI Flow Status⚛️ CI FlowRuleset - Version:
|
🔗 Helpful links
💊 CI failures summary and remediationsAs of commit 7591a35 (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. |
@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> Differential Revision: [D35010278](https://our.internmc.facebook.com/intern/diff/D35010278) [ghstack-poisoned]
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> ghstack-source-id: d0ea3286614051007c612f1f174d1715a0a5fba3 Pull Request resolved: #74468
@ezyang has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator. |
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> Differential Revision: [D35010278](https://our.internmc.facebook.com/intern/diff/D35010278) [ghstack-poisoned]
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> ghstack-source-id: 2442622f1e27092d7f55295208c64755435e3501 Pull Request resolved: #74468
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Stamped -- I look forward to MetaTestBase's return!
@pytorchbot merge this |
Hey @ezyang. |
Just catching up with this PR. Thanks a lot for making the change @ezyang! Porting remaining ops to structured kernels definitely sounds more appealing now :) |
After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> Pull Request resolved: #74468 Approved by: https://github.com/mruberry
Summary: After discussion with Can Balioglu, we have concluded that #53682 , while clever, is more trouble than it is worth. The main problem is that whenever someone adds support for new meta tensors, they then get dozens of new test case failures, because tests that were previously halted by lack of support for an operator on meta tensors, now have gotten further and hit some logic which expects to be able to, e.g., pull out a real value from a tensor (which clearly doesn't work). This is very annoying and time consuming! Most of these tests aren't written with meta device in mind, and it's not a good use of time to try to make them more generic. The plan on record is to switch meta testing to OpInfo, but that patch will take some time to prepare for now I want to stem the bleeding. I don't think we're at high risk for regressions here because meta tensors mostly share logic with their regular brethren. Signed-off-by: Edward Z. Yang <ezyangfb.com> Pull Request resolved: #74468 Approved by: https://github.com/mruberry Test Plan: contbuild & OSS CI, see https://hud.pytorch.org/commit/pytorch/pytorch/65329f4fac8fb22318b7a3eb115e9da207d8d41a Reviewed By: bigfootjon Differential Revision: D35193490 Pulled By: atalman fbshipit-source-id: 6e6bae34c075e60c6a84936f812e178bf88784cd
Stack from ghstack (oldest at bottom):
After discussion with Can Balioglu, we have concluded that
#53682 , while clever, is more
trouble than it is worth. The main problem is that whenever someone
adds support for new meta tensors, they then get dozens of new test case
failures, because tests that were previously halted by lack of support
for an operator on meta tensors, now have gotten further and hit some
logic which expects to be able to, e.g., pull out a real value from a
tensor (which clearly doesn't work). This is very annoying and time
consuming! Most of these tests aren't written with meta device in
mind, and it's not a good use of time to try to make them more generic.
The plan on record is to switch meta testing to OpInfo, but that patch
will take some time to prepare for now I want to stem the bleeding. I
don't think we're at high risk for regressions here because meta tensors
mostly share logic with their regular brethren.
Signed-off-by: Edward Z. Yang ezyang@fb.com
Differential Revision: D35010278