Skip to content

Wrong parameter order in all_to_all#65093

Merged
JZ-LIANG merged 1 commit intoPaddlePaddle:developfrom
eee4017:lawu/all_to_all
Jun 21, 2024
Merged

Wrong parameter order in all_to_all#65093
JZ-LIANG merged 1 commit intoPaddlePaddle:developfrom
eee4017:lawu/all_to_all

Conversation

@eee4017
Copy link
Contributor

@eee4017 eee4017 commented Jun 12, 2024

PR Category

Distributed Strategy

PR Types

Bug fixes

Description

The parameter order in PyBind differs from that in Python, causing a bug. By swapping the parameter order, this issue can be resolved.

Please refer to the PyBind C++ Interface (all_to_all_tensor_on_calc_stream and all_to_all_tensor.), where out_tensor should be the first parameter.

However, in the _all_to_all_tensor_in_dygraph, in_tensor is the first parameter.

Additionally, the parameter order in PaddlePaddle is not standardized, even in the highest-level APIs, which lack a consistent parameter sequence. Please check alltoall (stream) and alltoall (distributed)

@paddle-bot
Copy link

paddle-bot bot commented Jun 12, 2024

你的PR提交成功,感谢你对开源项目的贡献!
请关注后续CI自动化测试结果,详情请参考Paddle-CI手册
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

@paddle-bot paddle-bot bot added the contributor External developers label Jun 12, 2024
@jeng1220
Copy link
Collaborator

jeng1220 commented Jun 13, 2024

CI failed but it was NOT related to this PR

ERROR: test_simple_net_hybrid_strategy (__main__.TestSemiAutoParallelLlamaDataLoader)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/workspace/Paddle/test/collective/test_communication_api_base.py", line 79, in run_test_case
    self._launcher = subprocess.run(
  File "/usr/lib/python3.10/subprocess.py", line 526, in run
    raise CalledProcessError(retcode, process.args,
subprocess.CalledProcessError: Command '['/usr/bin/python', '-u', '-m', 'paddle.distributed.launch', '--log_dir', '/tmp/tmpkx_sv8w9', '--devices', '0,1,2,3,4,5,6,7', 'semi_auto_llama_dataloader.py']' returned non-zero exit status 1.
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
  File "/workspace/Paddle/test/auto_parallel/hybrid_strategy/test_semi_auto_parallel_llama_model.py", line 224, in test_simple_net_hybrid_strategy
    self.run_test_case(
  File "/workspace/Paddle/test/collective/test_communication_api_base.py", line 90, in run_test_case
    raise RuntimeError(
RuntimeError: Error occurs when running this test case. The return code of command ['/usr/bin/python', '-u', '-m', 'paddle.distributed.launch', '--log_dir', '/tmp/tmpkx_sv8w9', '--devices', '0,1,2,3,4,5,6,7', 'semi_auto_llama_dataloader.py'] is 1
----------------------------------------------------------------------
Ran 9 tests in 444.641s
FAILED (errors=1)

@onecatcn onecatcn assigned zyfncg and JZ-LIANG and unassigned zyfncg Jun 14, 2024
@tianshuo78520a
Copy link
Collaborator

这个问题还在修复中,修复后我会重新执行这个CI

Copy link
Contributor

@JZ-LIANG JZ-LIANG left a comment

Choose a reason for hiding this comment

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

后续最好补一个单测吧
test/collective/process_group_nccl.py

Copy link
Contributor

@JZ-LIANG JZ-LIANG left a comment

Choose a reason for hiding this comment

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

LGTM

@JZ-LIANG JZ-LIANG merged commit 5d3df27 into PaddlePaddle:develop Jun 21, 2024
co63oc pushed a commit to co63oc/Paddle that referenced this pull request Jun 25, 2024
Co-authored-by: lawrence910426 <lawu@nvidia.com>
co63oc pushed a commit to co63oc/Paddle that referenced this pull request Jun 25, 2024
Co-authored-by: lawrence910426 <lawu@nvidia.com>
deepllz added a commit to deepllz/Paddle that referenced this pull request Jul 4, 2024
sneaxiy pushed a commit that referenced this pull request Jul 4, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

contributor External developers NVIDIA

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants