frozenleaves/pytorch
1
0
Fork 0
mirror of https://github.com/pytorch/pytorch.git synced 2025-10-20 21:14:14 +08:00
Code Issues Packages Projects Releases Wiki Activity

inductor.config.descriptive_names = False is not actually supported (#145523)

Browse Source 
Summary:
This config is not supported (it throws an error when set), and doesn't really make sense imo.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/145523
Approved by: https://github.com/eellison
This commit is contained in:
Gabriel Ferns
2025-01-28 20:22:20 +00:00
committed by PyTorch MergeBot
parent 515e55e692
commit edf266e9bb
1 changed files with 2 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
torch/_inductor/config.py
View File

@ -874,7 +874,7 @@ class cpp:
# similar to config.triton.descriptive_names
descriptive_names: Union[
bool, Literal["torch", "original_aten", "inductor_node"]
Literal["torch", "original_aten", "inductor_node"]
] = "original_aten"
# how many nodes to allow into a single horizontal fusion
@ -1031,12 +1031,11 @@ class triton:
)
# should we put op names in kernel names
# False: No special names (just triton__1, triton__2, etc.)
# "torch": Maps to the fx op in the Dynamo graph (module name, method name, etc.)
# "original_aten": Maps to the highest-level aten op (i.e. pre-decompositions)
# "inductor_node": Maps to the node name in the FX graph passed to Inductor
descriptive_names: Union[
bool, Literal["torch", "original_aten", "inductor_node"]
Literal["torch", "original_aten", "inductor_node"]
] = "original_aten"
# use alternate codegen for smaller reductions