Summary:
Pull Request resolved: #135060
Current implementation for lifted graph takes a dict of [constant name: constant value]. And the constant value is used to run_node and excute the constant graph to get the folded values and then create new getattr nodes for folded values.
However, we don't have constant values for lifted graph during model compilation on MTIA. I think it makes the pass more general by allowing it to just take the constant names only to produce the constant graph and represent the folded nodes as placeholders to make it consistent with lifted graph. Additionally, this mimic the real situation in Sigmoid, where Sigmoid executes the constant graph, get the folded values and set the folded values to the main graph. This diff is to update the pass to work with a list of constant names.
Test Plan:
```
buck run mode/opt caffe2/test:test_export -- -r split_const_gm
```
Differential Revision: D62144791

Summary:
Pull Request resolved: pytorch#135060
Current implementation for lifted graph takes a dict of [constant name: constant value]. And the constant value is used to run_node and excute the constant graph to get the folded values and then create new getattr nodes for folded values.
However, we don't have constant values for lifted graph during model compilation on MTIA. I think it makes the pass more general by allowing it to just take the constant names only to produce the constant graph and represent the folded nodes as placeholders to make it consistent with lifted graph. Additionally, this mimic the real situation in Sigmoid, where Sigmoid executes the constant graph, get the folded values and set the folded values to the main graph. This diff is to update the pass to work with a list of constant names.
Test Plan:
```
buck run mode/opt caffe2/test:test_export -- -r split_const_gm
```
Differential Revision: D62144791

Summary:
Pull Request resolved: pytorch#135060
Current implementation for lifted graph takes a dict of [constant name: constant value]. And the constant value is used to run_node and excute the constant graph to get the folded values and then create new getattr nodes for folded values.
However, we don't have constant values for lifted graph during model compilation on MTIA. I think it makes the pass more general by allowing it to just take the constant names only to produce the constant graph and represent the folded nodes as placeholders to make it consistent with lifted graph. Additionally, this mimic the real situation in Sigmoid, where Sigmoid executes the constant graph, get the folded values and set the folded values to the main graph. This diff is to update the pass to work with a list of constant names.
Test Plan:
```
buck run mode/opt caffe2/test:test_export -- -r split_const_gm
```
Differential Revision: D62144791

Summary:
Pull Request resolved: pytorch#135060
Current implementation for lifted graph takes a dict of [constant name: constant value]. And the constant value is used to run_node and excute the constant graph to get the folded values and then create new getattr nodes for folded values.
However, we don't have constant values for lifted graph during model compilation on MTIA. I think it makes the pass more general by allowing it to just take the constant names only to produce the constant graph and represent the folded nodes as placeholders to make it consistent with lifted graph. Additionally, this mimic the real situation in Sigmoid, where Sigmoid executes the constant graph, get the folded values and set the folded values to the main graph. This diff is to update the pass to work with a list of constant names.
Test Plan:
```
buck run mode/opt caffe2/test:test_export -- -r split_const_gm
```
Reviewed By: SherlockNoMad
Differential Revision: D62144791

Summary:
Current implementation for lifted graph takes a dict of [constant name: constant value]. And the constant value is used to run_node and excute the constant graph to get the folded values and then create new getattr nodes for folded values.
However, we don't have constant values for lifted graph during model compilation on MTIA. I think it makes the pass more general by allowing it to just take the constant names only to produce the constant graph and represent the folded nodes as placeholders to make it consistent with lifted graph. Additionally, this mimic the real situation in Sigmoid, where Sigmoid executes the constant graph, get the folded values and set the folded values to the main graph. This diff is to update the pass to work with a list of constant names.
Test Plan:
```
buck run mode/opt caffe2/test:test_export -- -r split_const_gm
```
Reviewed By: muchulee8, SherlockNoMad
Differential Revision: D62144791

Summary:
Bug fix for #135060
Simple review:
https://github.com/pytorch/pytorch/pull/135060/files#diff-f23386709ff7e1235b15e18f835a48e5124e0ddd596aeb33c201daad1abbedd7R357
We mistakenly typed get_attr into getattr.
This causes constants never get untagged, and forces all constants get
cloned twice which greatly increases the memory consumption.
Test Plan:
python test/inductor/test_aot_inductor.py -k test_empty_constant_folding
Reviewers:
Subscribers:
Tasks:
Tags:
[ghstack-poisoned]

Summary:
Bug fix for #135060
Simple review:
https://github.com/pytorch/pytorch/pull/135060/files#diff-f23386709ff7e1235b15e18f835a48e5124e0ddd596aeb33c201daad1abbedd7R357
We mistakenly typed get_attr into getattr.
This causes constants never get untagged, and forces all constants get
cloned twice which greatly increases the memory consumption.
Test Plan:
python test/inductor/test_aot_inductor.py -k test_empty_constant_folding
Reviewers:
Subscribers:
Tasks:
Tags:
ghstack-source-id: 4ee4b54
Pull Request resolved: #152273

Summary:
Bug fix for #135060
Simple review:
https://github.com/pytorch/pytorch/pull/135060/files#diff-f23386709ff7e1235b15e18f835a48e5124e0ddd596aeb33c201daad1abbedd7R357
We mistakenly typed get_attr into getattr.
This causes constants never get untagged, and forces all constants get
cloned twice which greatly increases the memory consumption.
Test Plan:
python test/inductor/test_aot_inductor.py -k test_empty_constant_folding
Reviewers:
Subscribers:
Tasks:
Tags:
Pull Request resolved: #152273
Approved by: https://github.com/trieuat, https://github.com/zhxchen17