Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[fleet_executor] Parse pipeline config #37319

Merged
merged 5 commits into from
Nov 19, 2021

Conversation

FeixLiu
Copy link
Contributor

@FeixLiu FeixLiu commented Nov 18, 2021

PR types

Others

PR changes

Others

Describe

parse pipeline config.

@paddle-bot-old
Copy link

Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@FeixLiu FeixLiu marked this pull request as draft November 18, 2021 03:57
@FeixLiu FeixLiu marked this pull request as ready for review November 18, 2021 03:57
@PaddlePaddle PaddlePaddle locked and limited conversation to collaborators Nov 18, 2021
@PaddlePaddle PaddlePaddle unlocked this conversation Nov 18, 2021
@@ -27,4 +27,6 @@ message FleetExecutorDesc {
optional int32 dp_degree = 4 [ default = 1 ];
optional int32 mp_degree = 5 [ default = 1 ];
optional int32 pp_degree = 6 [ default = 1 ];
optional int64 num_micro_batches = 7 [ default = 1 ];
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

大概不是num_micro_steps?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个在python端就是global batch size / micro batch size,所以叫num_mircro_batches?一共有多少个mirco batch?其实就是num_micro_steps

@@ -27,4 +27,6 @@ message FleetExecutorDesc {
optional int32 dp_degree = 4 [ default = 1 ];
optional int32 mp_degree = 5 [ default = 1 ];
optional int32 pp_degree = 6 [ default = 1 ];
optional int64 num_micro_batches = 7 [ default = 1 ];
optional int64 num_slots = 8 [ default = 1 ];
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个是啥

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

就是一次最多能跑多少步

@@ -96,6 +96,9 @@ class Interceptor {
// local mailbox, written by FetchRemoteMailbox()
// read by PoolTheMailbox()
std::queue<InterceptorMessage> local_mailbox_;

int64_t already_run_times_{0};
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个建议加到后面的compute_interceptor中

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

这个是为了fake run准备的,留着吧,后面的子类可以不用?

if (role_to_ops.find(role_id) == role_to_ops.end()) {
task_nodes_.emplace_back(
TaskNode::CreateEmptyTaskNode(role_id, cur_rank, task_id));
task_nodes_.emplace_back(TaskNode::CreateEmptyTaskNode(
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

后续可能需要有ComputeTaskNode

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

没理解,为啥要单搞一个新的tasknode出来?

Copy link
Contributor

@wangxicoding wangxicoding left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM

@wangxicoding wangxicoding merged commit ca088f9 into PaddlePaddle:develop Nov 19, 2021
@FeixLiu FeixLiu deleted the parse_pipeline_config branch November 19, 2021 04:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants