Skip to content

Fix mixed precision output type to original type - #11142

Merged
comaniac merged 6 commits into
apache:mainfrom
gayatripk1:Fix_mixed_precision
May 5, 2022
Merged

Fix mixed precision output type to original type#11142
comaniac merged 6 commits into
apache:mainfrom
gayatripk1:Fix_mixed_precision

Conversation

@gayatripk1

@gayatripk1gayatripk1 commented Apr 27, 2022

Copy link
Copy Markdown
Contributor

Thanks for contributing to TVM! Please refer to guideline https://tvm.apache.org/docs/contribute/ for useful information and tips. After the pull request is submitted, please request code reviews from Reviewers by @ them in the pull request thread.

@gayatripk1

gayatripk1 commented Apr 27, 2022

Copy link
Copy Markdown
ContributorAuthor

cc @AndrewZhaoLuo@kparzysz-quic

@yangulei

yangulei commented Apr 27, 2022

Copy link
Copy Markdown
Contributor

Hi, does this mean the output OP has different dtype for inputs, weights and dst after AMP? If so, I'm afraid that my bfloat16 support for DNNL BYOC(#11111) might conflict with this. I implement bfloat16 support with the assumption that the dtype for the inputs, weights and dst are either all float32 or all bfloat16.

@comaniac

comaniac commented Apr 27, 2022

Copy link
Copy Markdown
Contributor

I think this PR is for the model output dtype only so it shouldn't affect other cases.

In addition to that, I'd suggest 1) exposing this behavior to users by adding a configure to PassContext, and 2) adding a unit test.

@yangulei

yangulei commented Apr 28, 2022

Copy link
Copy Markdown
Contributor

I think this PR is for the model output dtype only so it shouldn't affect other cases.

OK, thanks for your clarification.

@AndrewZhaoLuoAndrewZhaoLuo self-assigned this Apr 28, 2022
@AndrewZhaoLuo

AndrewZhaoLuo commented Apr 28, 2022

Copy link
Copy Markdown
Contributor

Sorry ive been busy, ill take a look tomorrow

@gayatripk1

gayatripk1 commented Apr 29, 2022

Copy link
Copy Markdown
ContributorAuthor

@AndrewZhaoLuo@comaniac, Modified this behavior to users by adding a configure to PassContext, and adding a unit test.

comaniac
comaniac approved these changes Apr 30, 2022
View reviewed changes

@comaniaccomaniac left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM. Just nits.

Comment threadsrc/relay/transforms/to_mixed_precision.cc Outdated
Comment threadsrc/relay/transforms/to_mixed_precision.cc Outdated
Comment threadtests/python/relay/test_to_mixed_precision.py Outdated
AndrewZhaoLuo
AndrewZhaoLuo approved these changes May 2, 2022
View reviewed changes

@AndrewZhaoLuoAndrewZhaoLuo left a comment •
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, thanks! Please fix CI though

@AndrewZhaoLuo

AndrewZhaoLuo commented May 3, 2022

Copy link
Copy Markdown
Contributor

Looks like CI is flaky, please push empty commit to restart CI. @gayatripk1

@gayatripk1

gayatripk1 commented May 4, 2022

Copy link
Copy Markdown
ContributorAuthor

Looks like CI is flaky, please push empty commit to restart CI. @gayatripk1

Done

@gayatripk1

gayatripk1 commented May 5, 2022

Copy link
Copy Markdown
ContributorAuthor

waiting for more reviews?

@comaniac
comaniac merged commit eae836c into apache:mainMay 5, 2022
@comaniac

comaniac commented May 5, 2022

Copy link
Copy Markdown
Contributor

Thanks @gayatripk1@AndrewZhaoLuo

shtinsa pushed a commit to Deelvin/tvm that referenced this pull request May 17, 2022
@gayatripk1Sergey Shtin
Fix mixed precision output type to original type (apache#11142)
37a6e64
SebastianBoblest pushed a commit to SebastianBoblest/tvm that referenced this pull request May 27, 2022
@gayatripk1Boblest Sebastian (ETAS-DEV/XPC-Fe1)
Fix mixed precision output type to original type (apache#11142)
9e46be9
Sign up for freeto join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@comaniaccomaniaccomaniac approved these changes
@AndrewZhaoLuoAndrewZhaoLuoAndrewZhaoLuo approved these changes

Assignees

@AndrewZhaoLuoAndrewZhaoLuo

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@gayatripk1@yangulei@comaniac@AndrewZhaoLuo