heter for collective #37613
Merged
heter for collective #37613
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
|
Thanks for your contribution! |
sandyhouse
reviewed
Nov 29, 2021
| @@ -176,6 +176,11 @@ void GLOOParallelContext::AllReduce(const framework::SelectedRows &src, | |||
| } | |||
| } | |||
|
|
|||
| void GLOOParallelContext::BroadCast(framework::Variable *src, int ring_id) { | |||
There was a problem hiding this comment.
Broadcast？Broadcast是一个单词。另外，这个接口没有实现，为什么还要添加这个接口呢？
| @@ -47,6 +47,8 @@ class GLOOParallelContext : public ParallelContext { | |||
| framework::Variable* dst, int ring_id, | |||
| bool use_calc_stream) override; | |||
|
|
|||
| void BroadCast(framework::Variable* src, int ring_id) override; | |||
| @@ -158,6 +158,29 @@ void HCCLParallelContext::AllReduceByStream(const framework::Variable &src, | |||
| } | |||
| } | |||
|
|
|||
| void HCCLParallelContext::BroadCast(framework::Variable *src, int ring_id) { | |||
| @@ -127,6 +135,20 @@ void NCCLParallelContext::AllReduceByStream(const framework::Variable &src, | |||
| AllReduce(src, dst, strategy_, ring_id, use_calc_stream); | |||
| } | |||
|
|
|||
| void NCCLParallelContext::BroadCast(framework::Variable *src, int ring_id) { | |||
| @@ -60,6 +60,8 @@ class NCCLParallelContext : public ParallelContext { | |||
| framework::Variable* dst, int ring_id, | |||
| bool use_calc_stream) override; | |||
|
|
|||
| void BroadCast(framework::Variable* src, int ring_id) override; | |||
| @@ -56,6 +56,8 @@ class ParallelContext { | |||
| framework::Variable* dst, int ring_id, | |||
| bool use_calc_stream) = 0; | |||
|
|
|||
| virtual void BroadCast(framework::Variable* src, int ring_id) = 0; | |||
| @@ -41,6 +42,9 @@ void Group::DivNRanks(const platform::DeviceContext &context, int64_t nranks) { | |||
| #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL) | |||
| DivNRanks(tensor, nranks, context); | |||
| #endif | |||
| } else if (platform::is_npu_place(tensor->place())) { | |||
| // TODO(kuizhiqing) | |||
| VLOG(4) << "divnrank for npu not support yet"; | |||
XieYunshen
approved these changes
Dec 1, 2021
chenwhql
approved these changes
Dec 1, 2021
fuyinno4
approved these changes
Dec 1, 2021
raindrops2sea
approved these changes
Dec 5, 2021
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
New features
PR changes
Others
Describe
Heterogenous mix training represents the model training with heterogenous hardwares. Dygraph mode is only supported now. GPU/NPU/XPU are targeting devices for this prototype work.
The basic idea is very similar as the use of hierarchical communication topology. The low layer reduce the data within each node, while the upper layer reduce across all global nodes.