add type promotion for complex and real number. #63842
Merged
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
|
你的PR提交成功，感谢你对开源项目的贡献! |
jeff41404
approved these changes
May 7, 2024
chenwhql
approved these changes
May 7, 2024
|
Sorry to inform you that 46238aa's CIs have passed for more than 7 days. To prevent PR conflicts, you need to re-run all CIs manually. |
co63oc
pushed a commit
to co63oc/Paddle
that referenced
this pull request
May 10, 2024
* add type promotion for complex and real number. * fix * reduce api support * add more api support * fix * fix * remove matmul * add T+S logic. * fix bug * fix unittest * fix * fix * fix unittest * fix gumbel * rm print * fix more unittests. * fix test_llama_group_log_softmax.py * fix bug, and add 0-d + 0-d logic. * rm print * fix behavior of bool and int * add unittest for all type promotion. * rm unintest which is unsupport dtype * fix * fix * add error unittest * fix increase unittest * bug fix * fixed by comment * remove useless code. * fix * fix * fix TypePromotionForZeroDimTensor * add inplace API support, add special case can skip type promotion (add x=float32,y=float16/bfloat16). * add broatcast support for MultiPrecisionAddKernelImpl.
co63oc
pushed a commit
to co63oc/Paddle
that referenced
this pull request
May 11, 2024
* add type promotion for complex and real number. * fix * reduce api support * add more api support * fix * fix * remove matmul * add T+S logic. * fix bug * fix unittest * fix * fix * fix unittest * fix gumbel * rm print * fix more unittests. * fix test_llama_group_log_softmax.py * fix bug, and add 0-d + 0-d logic. * rm print * fix behavior of bool and int * add unittest for all type promotion. * rm unintest which is unsupport dtype * fix * fix * add error unittest * fix increase unittest * bug fix * fixed by comment * remove useless code. * fix * fix * fix TypePromotionForZeroDimTensor * add inplace API support, add special case can skip type promotion (add x=float32,y=float16/bfloat16). * add broatcast support for MultiPrecisionAddKernelImpl.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
PR Category
Others
PR Types
New features
Description
card-78750
As there were unreasonable type promotion in Paddle, which the previous logic was aligned to the left tensor, like:
This behavior will be fixed more in line with mathematical logic, like:
Furthermore, after discussion, we will limit the behavior of automatic type promotion to floating-point numbers, and between real and complex numbers in Tensor and Tensor. Tensor and Scalar will still support all dtypes.
Those PR #60638 , #59518 fixed the behavior in few APIs between floating-point numbers in Tensor and Tensor.
This PR will support all binary operation API, and type promotion between real and complex numbers in Tensor and Tensor. Also the behavior between Tensor and Scalar will be corrected.