Fix inaccurate return of low precision op list #49391
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! |
AnnaTrainingG
force-pushed
the
fix_amp_lit
branch
2 times, most recently
from
f21b5ee to
f6faa79
Compare
zyfncg
reviewed
Jan 3, 2023
| @@ -1189,6 +1189,7 @@ def gen_kernel_code(self, kernel_name, code_indent, inplace_flag=False): | |||
| {code_indent} auto kernel_result = phi::KernelFactory::Instance().SelectKernelOrThrowError( | |||
| {code_indent} "{kernel_name}", {{kernel_backend, kernel_layout, kernel_data_type}}); | |||
| {code_indent} const auto& kernel = kernel_result.kernel; | |||
| {code_indent} phi::KernelFactory::Instance().AddToLowPrecisionKernelList("{self.api}", kernel_data_type); | |||
There was a problem hiding this comment.
感觉可以把 if 判断条件加在这一层以进一步减少额外的调用开销
| @@ -221,6 +221,7 @@ def gen_sparse_kernel_code(self, kernel_name, inplace_flag=False): | |||
| auto kernel_result = phi::KernelFactory::Instance().SelectKernelOrThrowError( | |||
| "{kernel_name}", {{kernel_backend, kernel_layout, kernel_data_type}}); | |||
| const auto& phi_kernel = kernel_result.kernel; | |||
| phi::KernelFactory::Instance().AddToLowPrecisionKernelList("{self.api}", kernel_data_type); | |||
| @@ -210,6 +210,7 @@ def gen_string_tensor_kernel_code(self, inplace_flag=False, code_indent=""): | |||
| VLOG(6) << "{self.api} api strings kernel key: [" << kernel_backend << ", " << kernel_layout << ", "<< kernel_data_type << "]"; | |||
| auto kernel_result = phi::KernelFactory::Instance().SelectKernelOrThrowError( | |||
| "{self.kernel['func'][0]}", {{kernel_backend, kernel_layout, kernel_data_type}}); | |||
| phi::KernelFactory::Instance().AddToLowPrecisionKernelList("{self.api}", kernel_data_type); | |||
AnnaTrainingG
force-pushed
the
fix_amp_lit
branch
from
f6faa79 to
9a5fc61
Compare
AnnaTrainingG
force-pushed
the
fix_amp_lit
branch
from
9a5fc61 to
6bf8c20
Compare
zyfncg
approved these changes
Jan 5, 2023
zkh2016
approved these changes
Jan 5, 2023
|
TODO： （2）不需要环境变量通过1、2这种等级控制，默认打印出来模型中所有算子的列表。在AMP任务中，一些算子可能是FP16或者FP32 Kernel都会调用 |
lanxianghit
approved these changes
Jan 6, 2023
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 types
Bug fixes
PR changes
Others
Describe
修复低精度算子列表获取不准确的问题。
说明：
整个功能通过FLAGS_low_precision_op_list环境变量控制，环境变量默认为0
在模型运行结束调用： paddle.amp.low_precision_op_list()
用法:
FLAGS_low_precision_op_list=1:返回当前模型前向低精度算子列表，便于低精度训练加入黑名单，注意对于inplace的OP无法通过加入黑名单的方式接触低精度运算。
FLAGS_low_precision_op_list=2:返回当前模型前向算子列表，便于查看算子列表。