[ONNX][TypePromo] Introduce ReductionTypePromotionRule #104491
Closed
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
[ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/104491
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 27ea2f3: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
Introduce `ReductionTypePromotionRule` and rename `TypePromotionRule` as
`ElementwiseTypePromotionRule`. Created base abstract class `TypePromotionRule`.
Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is
- "all", done
- "amax", done
- "amin", done
- "any", done
- "cumsum", done
- "cumprod", no torchlib impl
- "mean", done
- "std", no torchlib impl
- "std_mean", no torchlib impl
- "sum", done
- "sum_to_size", no torchlib impl
- "prod", no torchlib impl
- "var", no torchlib impl
- "var_mean", tricky. Node has multiple outputs. Follow up in separate PR.
[ghstack-poisoned]
…pePromo] Introduce ReductionTypePromotionRule"
Introduce `ReductionTypePromotionRule` and rename `TypePromotionRule` as
`ElementwiseTypePromotionRule`. Created base abstract class `TypePromotionRule`.
Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is
- "all", done
- "amax", done
- "amin", done
- "any", done
- "cumsum", done
- "cumprod", no torchlib impl
- "mean", done
- "std", no torchlib impl
- "std_mean", no torchlib impl
- "sum", done
- "sum_to_size", no torchlib impl
- "prod", no torchlib impl
- "var", no torchlib impl
- "var_mean", tricky. Node has multiple outputs. Follow up in separate PR.
[ghstack-poisoned]
…Introduce ReductionTypePromotionRule"
Introduce `ReductionTypePromotionRule` and rename `TypePromotionRule` as
`ElementwiseTypePromotionRule`. Created base abstract class `TypePromotionRule`.
Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is
- "all", done
- "amax", done
- "amin", done
- "any", done
- "cumsum", done
- "cumprod", no torchlib impl
- "mean", done
- "std", no torchlib impl
- "std_mean", no torchlib impl
- "sum", done
- "sum_to_size", no torchlib impl
- "prod", no torchlib impl
- "var", no torchlib impl
- "var_mean", tricky. Node has multiple outputs. Follow up in separate PR.
[ghstack-poisoned]
Introduce `ReductionTypePromotionRule` and rename `TypePromotionRule` as
`ElementwiseTypePromotionRule`. Created base abstract class `TypePromotionRule`.
Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is
- "all", done
- "amax", done
- "amin", done
- "any", done
- "cumsum", done
- "cumprod", no torchlib impl
- "mean", done
- "std", no torchlib impl
- "std_mean", no torchlib impl
- "sum", done
- "sum_to_size", no torchlib impl
- "prod", no torchlib impl
- "var", no torchlib impl
- "var_mean", tricky. Node has multiple outputs. Follow up in separate PR.
[ghstack-poisoned]
Introduce `ReductionTypePromotionRule` and rename `TypePromotionRule` as
`ElementwiseTypePromotionRule`. Created base abstract class `TypePromotionRule`.
Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is
- "all", done
- "amax", done
- "amin", done
- "any", done
- "cumsum", done
- "cumprod", no torchlib impl
- "mean", done
- "std", no torchlib impl
- "std_mean", no torchlib impl
- "sum", done
- "sum_to_size", no torchlib impl
- "prod", no torchlib impl
- "var", no torchlib impl
- "var_mean", tricky. Node has multiple outputs. Follow up in separate PR.
[ghstack-poisoned]
justinchuby
approved these changes
Jul 12, 2023
thiagocrepaldi
approved these changes
Jul 12, 2023
| "unflatten", | ||
| # "var_mean", # Segfault during onnx shape inference. Need to bump onnx version. |
There was a problem hiding this comment.
Is there a issue to track it? or at least add a TODO:
There was a problem hiding this comment.
Sure #105073
Stack from ghstack (oldest at bottom):
Introduce
ReductionTypePromotionRuleand renameTypePromotionRuleasElementwiseTypePromotionRule. Created base abstract classTypePromotionRule.Reduction rules are manually curated because the total number of ops is low, yet
most of them require some special treatment. The list that are covered in our unittest is