[Auto Parallel]: Support std::vector<phi::Tensor> input and output for DistTensor. #56602
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
Concat forward and backward are verified.
chenwhql
reviewed
Aug 24, 2023
… dist_tensor_support
… dist_tensor_support
…Tensor> -> std::vector<Tensor>.
… dist_tensor_support
…or output of operators. Following testcases are passed. 1. concat: std::vector<phi::Tensor> -> phi::Tensor 2. unbind: phi::Tensor -> std::vector<phi::Tensor> 3. broadcast_tensors: std::vector<phi::Tensor> -> std::vector<phi::Tensor>
GhostScreaming
changed the title
GhostScreaming
changed the title
chenwhql
reviewed
Sep 4, 2023
| phi::distributed::DistTensor* dist_tensor = | ||
| static_cast<phi::distributed::DistTensor*>(tensor.impl().get()); | ||
| intermidiate_tensor_.set_impl( | ||
| std::make_shared<phi::distributed::DistTensor>( |
There was a problem hiding this comment.
这里新构造tensor会触发重切分，我们还需要讨论一下，这样能过的话，也可以先这样写
There was a problem hiding this comment.
目前在concat上验证，是没有问题的。可以先加一个TODO？
LiYuRio
approved these changes
Sep 5, 2023
| const TransformFlag& transform_flag, | ||
| bool is_stride_kernel) { | ||
| std::vector<std::shared_ptr<phi::distributed::DistTensor>> out; | ||
| for (auto x : input) { |
| dense_tensor.meta().is_contiguous()))) { | ||
| out.push_back( | ||
| std::static_pointer_cast<phi::distributed::DistTensor>(tensor_in)); | ||
| continue; |
There was a problem hiding this comment.
理论上这里可以写到else分支里，不用continue
chenwhql
approved these changes
Sep 5, 2023
BeingGod
pushed a commit
to BeingGod/Paddle
that referenced
this pull request
Sep 9, 2023
…r DistTensor. (PaddlePaddle#56602) * [WIP] Support std::vector<phi::Tensor> input and output for DistTensor. Concat forward and backward are verified. * Polish code for new dist tensor implementation. * Fix bug of DistTensor upgrade. Add support functions for std::vector<Tensor> -> std::vector<Tensor>. * Add support for DistTensor type of std::vector<phi::Tensor> as input or output of operators. Following testcases are passed. 1. concat: std::vector<phi::Tensor> -> phi::Tensor 2. unbind: phi::Tensor -> std::vector<phi::Tensor> 3. broadcast_tensors: std::vector<phi::Tensor> -> std::vector<phi::Tensor> * Polish code. Remove useless comments. * Add update_loss_scaling in skip_op_lists. * Polish code.
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
Pcard-73145
Support
std::vector<phi::Tensor>input and output forDistTensor. Meanwhileno_need_buffertype is also supported.Concat,Broadcast_Tensors,Unbindforward and backward are verified. Following operators need to be supported later (along with their backward op if exists):check_finite_and_unscale,coalesce_tensor,meshgrid,update_loss_scaling,einsum.Foward operators have output of
std::tuple<...>are not supported now.