[quant][graphmode] Remove dedup pass#39825
[quant][graphmode] Remove dedup pass#39825jerryzh168 wants to merge 8 commits intogh/jerryzh168/339/basefrom
Conversation
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: [ghstack-poisoned]
💊 CI failures summary and remediationsAs of commit 980ad71 (more details on the Dr. CI page):
ci.pytorch.org: 1 failedThis comment was automatically generated by Dr. CI (expand for details).Follow this link to opt-out of these comments for your Pull Requests.Please report bugs/suggestions on the GitHub issue tracker or post in the (internal) Dr. CI Users group. This comment has been revised 21 times. |
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
| if not all(isinstance(x, str) for x in qconfig_dict.keys()): | ||
| raise ValueError('qconfig_dict should only contain names(str) as keys.') | ||
| scripted_qconfig_dict = script_qconfig_dict(qconfig_dict) | ||
| torch._C._jit_pass_dedup_module_uses(model._c) |
There was a problem hiding this comment.
Dont you need to remove the corresponding tests?
I thought this pass was needed to handle cases like:
- x = self.conv1(x)
- x = self.relu(x)
- x = self.conv2(x)
- x = self.relu(x)
There was a problem hiding this comment.
this is a legacy pass we added for this case, but this case can be addressed with current code already. I feel this pass might not be needed if we have some (unlikely) case like:
x = self.submodule_with_states(x)
...
x = self.submodule_with_states(x)that is: we use some submodule with states multiple times.
I only removed this from API, this pass is still there. If we decide to support the above use case at some point, we'll debug the problem and re-enable the pass.
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
Summary: Removing the pass for now since it is causing error for some models Test Plan: Reviewers: Subscribers: Tasks: Tags: Differential Revision: [D21987878](https://our.internmc.facebook.com/intern/diff/D21987878) [ghstack-poisoned]
|
This pull request has been merged in 14e841c. |
Summary: Pull Request resolved: pytorch#39825 Removing the pass for now since it is causing error for some models Test Plan: Imported from OSS Differential Revision: D21987878 fbshipit-source-id: 129aefb34754d5390a4c9d3108fa1b6c2eae5a74
Summary: Pull Request resolved: pytorch#39825 Removing the pass for now since it is causing error for some models Test Plan: Imported from OSS Differential Revision: D21987878 fbshipit-source-id: 129aefb34754d5390a4c9d3108fa1b6c2eae5a74
Stack from ghstack:
Summary:
Removing the pass for now since it is causing error for some models
Test Plan:
Reviewers:
Subscribers:
Tasks:
Tags:
Differential Revision: D21987878