[dy2s] fix error when using same tensor as inputs in one call & fix bugs in jit.save #55963
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! |
Aurelius84
reviewed
Aug 7, 2023
| @@ -321,7 +346,8 @@ inline void RunProgramAPI( | |||
|
|
|||
| VLOG(4) << "global_inner_scope:" << global_inner_scope; | |||
|
|
|||
| auto input_names = details::GetTensorsName(x); | |||
| auto input_names = | |||
There was a problem hiding this comment.
@feifei-111 之前的方案会有什么问题？Tensor.name 不符合预期么？
There was a problem hiding this comment.
如果两个输入是同一个tensor，组的program会不一致，这个导致sot里面没办法复用之前的program，因为sot希望能直接绕过动转静的cache拿到program。现在会对输入的tensor重命名，这样生成的program就没问题了
There was a problem hiding this comment.
原来的Tensor.name无法处理同一个Tensor绑定了Program中两个名字的情况。现在是可以兼容所有的情况的。
jiweibo
previously approved these changes
Aug 7, 2023
feifei-111
changed the title
… fix_run_program_op
… fix_run_program_op
2742195759
approved these changes
Aug 18, 2023
Comment on lines
+152
to
+176
| static void ShareTensorsIntoScopeWithName( | ||
| const std::vector<Tensor> &tensors, | ||
| const std::vector<std::string> &tensor_names, | ||
| paddle::framework::Scope *scope) { | ||
| for (size_t i = 0; i < tensors.size(); ++i) { | ||
| auto name = tensor_names[i]; | ||
| if (name == paddle::framework::kFakeVarName) { | ||
| continue; | ||
| } | ||
| auto *var = scope->Var(name); | ||
| CheckInputVarStatus(tensors[i]); | ||
| // share tensor | ||
| auto tensor_base = tensors[i].impl(); | ||
| if (phi::DenseTensor::classof(tensor_base.get())) { | ||
| auto *dst_tensor = var->GetMutable<phi::DenseTensor>(); | ||
| auto t = std::dynamic_pointer_cast<phi::DenseTensor>(tensor_base); | ||
| *dst_tensor = *t; | ||
| } else if (phi::SelectedRows::classof(tensor_base.get())) { | ||
| auto *dst_tensor = var->GetMutable<phi::SelectedRows>(); | ||
| auto t = std::dynamic_pointer_cast<phi::SelectedRows>(tensor_base); | ||
| *dst_tensor = *t; | ||
| } | ||
| } | ||
| } | ||
|
|
There was a problem hiding this comment.
消除重复，这个可以被 ShareTensorsIntoScope 复用。
| @@ -488,11 +491,12 @@ def keep_name_table(self, value): | |||
|
|
|||
| def _parse_save_configs(configs): | |||
| @@ -302,6 +311,44 @@ def _replace_value_with_input_spec(args): | |||
| return args_with_spec | |||
|
|
|||
|
|
|||
| def _replace_to_input_spec_with_new_name(args, arg_names): | |||
There was a problem hiding this comment.
删除掉没用的：_replace_value_with_input_spec
| @@ -887,6 +904,8 @@ def _prepare(self, inputs): | |||
| flatten_inputs = paddle.utils.flatten(inputs) | |||
| # Convert variable into Tensor and feed in training data. | |||
| input_vars = [] | |||
| input_var_names = [] | |||
There was a problem hiding this comment.
写成函数式：
map(lambda x: x.desct.name, self.inputs)
YuanRisheng
approved these changes
Aug 18, 2023
vivienfanghuagood
approved these changes
Aug 21, 2023
heavengate
approved these changes
Aug 21, 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
Others
PR changes
Others
Description
fix error when using same tensor as inputs in one call
---- wrong program created, when case like:
t = paddle.to_tensor(1)
f(t, t)
fix bugs in jit.save, when program_cache is not empty, jit.save might save a wrong program when input_spec is given
Others
PCard-66972