This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
use safe accumulation for norm #14240
Closed
Conversation
This file contains 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
|
@mxnet-label-bot add[Backend, Operator, pr-awaiting-review] |
marcoabreu
added
Backend
|
I did not see it immediately from the code - what is the type of the result? Norm should output fp32 as it's result since it is very easy to overflow otherwise. Mean on the other hand can be done with just fp32 accumulator without changing the output type. |
|
I didn't change the output type yet. Looks like I should add a dtype option to norm too. |
| "floating point types not int8"; \ | ||
| { \ | ||
| typedef int8_t DType; \ | ||
| typedef int8_t AType; \ |
There was a problem hiding this comment.
can we support acc in int types, too?
szha
added
pr-work-in-progress
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
use safe accumulation for norm. resolves #14126
Checklist
Essentials
Please feel free to remove inapplicable items for your PR.
Changes