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.

Sign up for GitHub

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

Jump to bottom

Add runtime error info in eager boxing #7926

Merged
merged 21 commits into from
May 31, 2022
Merged

Add runtime error info in eager boxing #7926

merged 21 commits into from
May 31, 2022

Conversation

clackhan
Copy link
Contributor

添加eager boxing 运行时错误信息

@clackhan clackhan requested review from hjchen2 and wyg1997 March 30, 2022 03:15
@clackhan clackhan requested a review from oneflow-ci-bot May 27, 2022 02:34
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.1ms (= 13012.0ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.9ms (= 14386.4ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 143.9ms / 130.1ms)

OneFlow resnet50 time: 76.8ms (= 7680.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 87.6ms (= 8756.5ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.14 (= 87.6ms / 76.8ms)

OneFlow resnet50 time: 53.8ms (= 10765.3ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 53.8ms (= 10756.4ms / 200, input_shape=[4, 3, 224, 224])
❌ Relative speed: 1.00 (= 53.8ms / 53.8ms)

OneFlow resnet50 time: 39.9ms (= 7978.6ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 44.4ms (= 8874.3ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.11 (= 44.4ms / 39.9ms)

OneFlow resnet50 time: 35.6ms (= 7113.6ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.2ms (= 7831.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.10 (= 39.2ms / 35.6ms)

OneFlow swin dataloader time: 0.251s (= 50.250s / 200, num_workers=1)
PyTorch swin dataloader time: 0.151s (= 30.117s / 200, num_workers=1)
Relative speed: 0.599 (= 0.151s / 0.251s)

OneFlow swin dataloader time: 0.113s (= 22.612s / 200, num_workers=4)
PyTorch swin dataloader time: 0.043s (= 8.586s / 200, num_workers=4)
Relative speed: 0.380 (= 0.043s / 0.113s)

OneFlow swin dataloader time: 0.037s (= 7.414s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.661s / 200, num_workers=8)
Relative speed: 0.629 (= 0.023s / 0.037s)

❌ OneFlow resnet50 time: 145.7ms (= 14568.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 172.1ms (= 17211.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.18 (= 172.1ms / 145.7ms)

OneFlow resnet50 time: 96.6ms (= 9655.4ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 112.8ms (= 11279.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.17 (= 112.8ms / 96.6ms)

OneFlow resnet50 time: 73.5ms (= 14698.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 87.6ms (= 17527.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.19 (= 87.6ms / 73.5ms)

OneFlow resnet50 time: 60.4ms (= 12080.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.1ms (= 15215.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 76.1ms / 60.4ms)

OneFlow resnet50 time: 55.6ms (= 11126.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.5ms (= 14496.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.30 (= 72.5ms / 55.6ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/7926/

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.4ms (= 13036.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 144.5ms (= 14446.1ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 144.5ms / 130.4ms)

OneFlow resnet50 time: 77.7ms (= 7771.8ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 87.9ms (= 8785.9ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.13 (= 87.9ms / 77.7ms)

OneFlow resnet50 time: 55.0ms (= 10993.9ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 59.1ms (= 11810.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.07 (= 59.1ms / 55.0ms)

OneFlow resnet50 time: 42.0ms (= 8401.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 48.8ms (= 9768.4ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.16 (= 48.8ms / 42.0ms)

OneFlow resnet50 time: 35.7ms (= 7141.6ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 37.5ms (= 7504.2ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.05 (= 37.5ms / 35.7ms)

OneFlow swin dataloader time: 0.380s (= 76.063s / 200, num_workers=1)
PyTorch swin dataloader time: 0.150s (= 30.048s / 200, num_workers=1)
Relative speed: 0.395 (= 0.150s / 0.380s)

OneFlow swin dataloader time: 0.065s (= 12.937s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.430s / 200, num_workers=4)
Relative speed: 0.652 (= 0.042s / 0.065s)

OneFlow swin dataloader time: 0.035s (= 7.080s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.447s / 200, num_workers=8)
Relative speed: 0.628 (= 0.022s / 0.035s)

❌ OneFlow resnet50 time: 146.9ms (= 14687.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 173.7ms (= 17366.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.18 (= 173.7ms / 146.9ms)

OneFlow resnet50 time: 100.8ms (= 10080.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 113.3ms (= 11332.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 113.3ms / 100.8ms)

OneFlow resnet50 time: 74.2ms (= 14845.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 94.2ms (= 18840.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 94.2ms / 74.2ms)

OneFlow resnet50 time: 62.5ms (= 12492.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 77.3ms (= 15455.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 77.3ms / 62.5ms)

OneFlow resnet50 time: 55.2ms (= 11037.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 80.2ms (= 16047.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.45 (= 80.2ms / 55.2ms)

@clackhan clackhan requested review from oneflow-ci-bot and removed request for oneflow-ci-bot May 27, 2022 07:43
wyg1997
wyg1997 reviewed May 27, 2022
View reviewed changes
oneflow/core/boxing/asymmetric_broadcast.cpp Outdated
Comment on lines 40 to 45
CHECK_EQ_OR_RETURN(in->nd_sbp()->sbp_parallel_size(), 1); // NOLINT(maybe-need-error-msg)
CHECK_EQ_OR_RETURN(out->nd_sbp()->sbp_parallel_size(), 1); // NOLINT(maybe-need-error-msg)
CHECK_OR_RETURN(IsAllBroadcastNdSbp(in->nd_sbp())); // NOLINT(maybe-need-error-msg)
CHECK_OR_RETURN(IsAllBroadcastNdSbp(out->nd_sbp())); // NOLINT(maybe-need-error-msg)
CHECK_OR_RETURN(out->placement()->Bigger(*in->placement()) // NOLINT(maybe-need-error-msg)
|| in->placement()->Bigger(*out->placement())); // NOLINT(maybe-need-error-msg)
Copy link
Contributor

Choose a reason for hiding this comment

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

我们的 clang-tidy 版本更新了，可以用 // NOLINTBEGIN(maybe-need-error-msg) 了，你这里也直接用吧，要不我一直要处理冲突 🤣

Copy link
Contributor Author

Choose a reason for hiding this comment

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

我们的 clang-tidy 版本更新了，可以用 // NOLINTBEGIN(maybe-need-error-msg) 了，你这里也直接用吧，要不我一直要处理冲突 🤣

好的👌

Copy link
Contributor Author

Choose a reason for hiding this comment

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

我们的 clang-tidy 版本更新了，可以用 // NOLINTBEGIN(maybe-need-error-msg) 了，你这里也直接用吧，要不我一直要处理冲突 🤣

已修改

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/7926/

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/7926/

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce GTX 1080 

❌ OneFlow resnet50 time: 130.9ms (= 13085.1ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.9ms (= 14394.5ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.10 (= 143.9ms / 130.9ms)

OneFlow resnet50 time: 78.6ms (= 7856.7ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 86.4ms (= 8636.3ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.10 (= 86.4ms / 78.6ms)

OneFlow resnet50 time: 55.1ms (= 11023.0ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.5ms (= 12097.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.10 (= 60.5ms / 55.1ms)

OneFlow resnet50 time: 42.5ms (= 8502.5ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 43.4ms (= 8673.8ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.02 (= 43.4ms / 42.5ms)

OneFlow resnet50 time: 35.8ms (= 7161.9ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 37.9ms (= 7570.1ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.06 (= 37.9ms / 35.8ms)

OneFlow swin dataloader time: 0.240s (= 47.912s / 200, num_workers=1)
PyTorch swin dataloader time: 0.149s (= 29.781s / 200, num_workers=1)
Relative speed: 0.622 (= 0.149s / 0.240s)

OneFlow swin dataloader time: 0.066s (= 13.121s / 200, num_workers=4)
PyTorch swin dataloader time: 0.040s (= 8.078s / 200, num_workers=4)
Relative speed: 0.616 (= 0.040s / 0.066s)

OneFlow swin dataloader time: 0.041s (= 8.218s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.382s / 200, num_workers=8)
Relative speed: 0.533 (= 0.022s / 0.041s)

❌ OneFlow resnet50 time: 146.2ms (= 14623.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 175.1ms (= 17508.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 175.1ms / 146.2ms)

OneFlow resnet50 time: 97.1ms (= 9709.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 112.1ms (= 11212.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 112.1ms / 97.1ms)

OneFlow resnet50 time: 71.9ms (= 14372.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 88.1ms (= 17611.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.23 (= 88.1ms / 71.9ms)

OneFlow resnet50 time: 61.2ms (= 12247.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.4ms (= 14875.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.21 (= 74.4ms / 61.2ms)

OneFlow resnet50 time: 55.5ms (= 11100.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.7ms (= 13747.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 68.7ms / 55.5ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/7926/

@clackhan clackhan merged commit a20b207 into master May 31, 2022
@clackhan clackhan deleted the add_runtime_error_info_in_eager_boxing branch May 31, 2022 07:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@hjchen2 hjchen2 hjchen2 approved these changes

@wyg1997 wyg1997 wyg1997 approved these changes

@oneflow-ci-bot oneflow-ci-bot Awaiting requested review from oneflow-ci-bot

Assignees
No one assigned
Labels
automerge enhancement need-check-error-message system
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@clackhan @hjchen2 @wyg1997 @zhongshsh