feat: add pipeline parallel #88
Open
Conversation
This file contains hidden or 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
JYMiracle305
force-pushed
the
add_pp
branch
2 times, most recently
from
7552d8e to
1083190
Compare
JYMiracle305
force-pushed
the
add_pp
branch
4 times, most recently
from
9c47bc2 to
4e47615
Compare
kilinchange
requested changes
Nov 13, 2025
JYMiracle305
force-pushed
the
add_pp
branch
6 times, most recently
from
09735f2 to
c21cd75
Compare
Contributor
Author
|
GPT2: 单卡： 
PP Num_micro_batch 1 
PP Num_micro_batch 2 
PP Num_micro_batch 4 
PP Num_micro_batch 8 
|
Contributor
Author
|
LLaMA3： 单卡： 
PP Num_micro_batch 1：
PP Num_micro_batch 2： 
PP Num_micro_batch 4： PP Num_micro_batch 8：
|
kilinchange
requested changes
Nov 17, 2025
kilinchange
approved these changes
Nov 17, 2025
JYMiracle305
force-pushed
the
add_pp
branch
3 times, most recently
from
9354e7a to
47e0573
Compare
Chamberlain0w0
requested changes
Nov 17, 2025
|
|
||
| void SGD::Step() { | ||
| for (auto param : params_) { | ||
| if (!param->grad()) { |
Contributor
There was a problem hiding this comment.
这个情况会出现吗？如果要做非空检查的话是不是后面的 Adam::Step 也得加上？
| std::vector<std::shared_ptr<Tensor>> target_mbs(num_micro_batches_); | ||
| if (stage_->IsFirstStage()) { | ||
| { | ||
| autograd::NoGradGuard no_grad; |
Contributor
There was a problem hiding this comment.
check 一下 no_grad 是否有必要，全模型的输入输出 input/target 的 requires_grad 是 false，相关操作不会构图
|
|
||
| if (stage_->IsLastStage()) { | ||
| { | ||
| autograd::NoGradGuard no_grad; |
|
|
||
| std::vector<std::shared_ptr<Tensor>> outputs; | ||
| for (auto t : input_tensors) { outputs.push_back(t); } | ||
| return outputs; |
Contributor
There was a problem hiding this comment.
这个是不是可以原地返回 input_tensors，没有必要再构造一个 outputs
JYMiracle305
force-pushed
the
add_pp
branch
2 times, most recently
from
3391267 to
b851306
Compare
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.
1. 主要修改：
新增支持Pipeline Parallel（PP）特性，支持把原始model按最外层结构平均分配到多个rank，各个rank持有部分layer以及相应parameters，训练过程forward按计算图顺序从rank 0向rank n计算，barkward从rank n向rank 0计算，各个rank之间通过点对点通信。
net.cc： 根据PP_size和pp_rank在构建模型时构建属于本rank的子块和对应参数。
pipeline_parallel.cc： PipelineParallel封装model，每个rank对应1个PipelineParallel，完成关联PipelineSchedule、PipelineStage、Optimizers，提供TrainStep函数为训练入口，调用调度器的训练方法Step。
pipeline_schedule.cc： PipelineSchedule调度器基类，提供Step函数为完整一轮训练的方法；ScheduleGPipe为调度器子类GPipe实现（示意图如下），StepMicrobatches为调度具体实现。
pipeline_stage.cc： PipelineStage表示当前rank所持有的子图，提供ForwardOneChunk方法执行当前子图内部的forward的计算。
send_recv.cc：ISend和IRecv是两个autograd节点，用于在rank间定向发送张量，依赖于autograd机制，
在rank x的反向中最后一步为发送梯度到rank x-1，然后调用rank x-1上的ISend::Backward接收梯度，并开始rank x-1的反向过程。2. 命令参数：
--pipeline_parallel #uint32类型，表示打开pipeline parallel，参数值为并行的设备数，即stage数量
示例：
./llama3 --input_bin <input_path> --llmc_filepath <model_path> --device cuda --nthread_per_process 8 --batch_size 10 --total_batch_size 5120 --num_iteration 10 --pipeline_parallel 8