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

eager global min_max oberver #7725

Merged
merged 11 commits into from
Mar 30, 2022
Merged

eager global min_max oberver #7725

merged 11 commits into from
Mar 30, 2022

Conversation

liufengwei0103
Copy link
Contributor

No description provided.

@github-actions
Copy link
Contributor

CI failed when running job: cuda-module-distributed-rank-0. PR label automerge has been removed

@github-actions
Copy link
Contributor

CI failed when running job: cuda-module-distributed-rank-1. PR label automerge has been removed

@liufengwei0103 liufengwei0103 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot March 29, 2022 07:50
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.4ms (= 12839.9ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.0ms (= 14097.3ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.10 (= 141.0ms / 128.4ms)

✔️ OneFlow resnet50 time: 78.8ms (= 7876.5ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 81.0ms (= 8104.9ms / 100, input_shape=[8, 3, 224, 224])
❌ Relative speed: 1.03 (= 81.0ms / 78.8ms)

OneFlow resnet50 time: 51.9ms (= 10375.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 54.3ms (= 10861.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.05 (= 54.3ms / 51.9ms)

OneFlow resnet50 time: 43.5ms (= 8705.6ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.2ms (= 9833.4ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.13 (= 49.2ms / 43.5ms)

OneFlow resnet50 time: 40.0ms (= 8005.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.6ms (= 7713.5ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 0.96 (= 38.6ms / 40.0ms)

OneFlow swin dataloader time: 0.247s (= 49.499s / 200, num_workers=1)
PyTorch swin dataloader time: 0.256s (= 51.276s / 200, num_workers=1)
✔️ Relative speed: 1.036 (= 0.256s / 0.247s)

OneFlow swin dataloader time: 0.067s (= 13.445s / 200, num_workers=4)
PyTorch swin dataloader time: 0.082s (= 16.416s / 200, num_workers=4)
✔️ Relative speed: 1.221 (= 0.082s / 0.067s)

OneFlow swin dataloader time: 0.037s (= 7.345s / 200, num_workers=8)
PyTorch swin dataloader time: 0.037s (= 7.407s / 200, num_workers=8)
✔️ Relative speed: 1.008 (= 0.037s / 0.037s)

✔️ OneFlow resnet50 time: 135.3ms (= 13532.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 157.2ms (= 15722.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 157.2ms / 135.3ms)

OneFlow resnet50 time: 89.3ms (= 8930.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 100.0ms (= 9997.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 100.0ms / 89.3ms)

OneFlow resnet50 time: 60.9ms (= 12183.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 76.6ms (= 15322.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 76.6ms / 60.9ms)

OneFlow resnet50 time: 53.0ms (= 10598.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.5ms (= 13292.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 66.5ms / 53.0ms)

OneFlow resnet50 time: 50.2ms (= 10046.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 61.5ms (= 12293.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 61.5ms / 50.2ms)

@github-actions
Copy link
Contributor

CI failed when running job: cuda-speed-test. PR label automerge has been removed

@liufengwei0103
Copy link
Contributor Author

test case code without affecting speed

@liufengwei0103 liufengwei0103 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot March 29, 2022 14:20
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.3ms (= 12828.1ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.8ms (= 14180.3ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.11 (= 141.8ms / 128.3ms)

✔️ OneFlow resnet50 time: 77.2ms (= 7718.2ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.6ms (= 8359.3ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.08 (= 83.6ms / 77.2ms)

OneFlow resnet50 time: 52.3ms (= 10451.8ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.2ms (= 11643.2ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.11 (= 58.2ms / 52.3ms)

OneFlow resnet50 time: 41.5ms (= 8294.5ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 46.5ms (= 9291.7ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.12 (= 46.5ms / 41.5ms)

OneFlow resnet50 time: 38.5ms (= 7695.9ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.8ms (= 8168.1ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.06 (= 40.8ms / 38.5ms)

OneFlow swin dataloader time: 0.249s (= 49.791s / 200, num_workers=1)
PyTorch swin dataloader time: 0.252s (= 50.429s / 200, num_workers=1)
✔️ Relative speed: 1.013 (= 0.252s / 0.249s)

OneFlow swin dataloader time: 0.066s (= 13.226s / 200, num_workers=4)
PyTorch swin dataloader time: 0.066s (= 13.228s / 200, num_workers=4)
✔️ Relative speed: 1.000 (= 0.066s / 0.066s)

OneFlow swin dataloader time: 0.036s (= 7.284s / 200, num_workers=8)
PyTorch swin dataloader time: 0.037s (= 7.328s / 200, num_workers=8)
✔️ Relative speed: 1.006 (= 0.037s / 0.036s)

✔️ OneFlow resnet50 time: 135.3ms (= 13531.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 156.3ms (= 15628.7ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 156.3ms / 135.3ms)

OneFlow resnet50 time: 89.2ms (= 8918.6ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 100.2ms (= 10015.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 100.2ms / 89.2ms)

OneFlow resnet50 time: 62.1ms (= 12413.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.5ms (= 15693.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 78.5ms / 62.1ms)

OneFlow resnet50 time: 51.1ms (= 10215.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.7ms (= 15138.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.48 (= 75.7ms / 51.1ms)

OneFlow resnet50 time: 48.5ms (= 9706.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 61.9ms (= 12383.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.28 (= 61.9ms / 48.5ms)

@github-actions
Copy link
Contributor

CI failed when running job: cuda-module-distributed-rank-1. PR label automerge has been removed

@liufengwei0103 liufengwei0103 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot March 30, 2022 00:40
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.3ms (= 12830.4ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 140.3ms (= 14032.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.09 (= 140.3ms / 128.3ms)

✔️ OneFlow resnet50 time: 77.7ms (= 7770.8ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.6ms (= 8357.1ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.08 (= 83.6ms / 77.7ms)

OneFlow resnet50 time: 49.8ms (= 9969.0ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 61.9ms (= 12385.2ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.24 (= 61.9ms / 49.8ms)

OneFlow resnet50 time: 43.0ms (= 8601.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 46.9ms (= 9388.7ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.09 (= 46.9ms / 43.0ms)

OneFlow resnet50 time: 36.3ms (= 7268.9ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.1ms (= 8024.4ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.10 (= 40.1ms / 36.3ms)

OneFlow swin dataloader time: 0.245s (= 48.991s / 200, num_workers=1)
PyTorch swin dataloader time: 0.250s (= 49.920s / 200, num_workers=1)
✔️ Relative speed: 1.019 (= 0.250s / 0.245s)

OneFlow swin dataloader time: 0.066s (= 13.154s / 200, num_workers=4)
PyTorch swin dataloader time: 0.067s (= 13.409s / 200, num_workers=4)
✔️ Relative speed: 1.019 (= 0.067s / 0.066s)

OneFlow swin dataloader time: 0.037s (= 7.406s / 200, num_workers=8)
PyTorch swin dataloader time: 0.037s (= 7.439s / 200, num_workers=8)
✔️ Relative speed: 1.004 (= 0.037s / 0.037s)

✔️ OneFlow resnet50 time: 135.4ms (= 13543.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 156.4ms (= 15642.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 156.4ms / 135.4ms)

OneFlow resnet50 time: 87.0ms (= 8704.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 106.4ms (= 10639.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 106.4ms / 87.0ms)

OneFlow resnet50 time: 59.0ms (= 11805.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 77.0ms (= 15399.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.30 (= 77.0ms / 59.0ms)

OneFlow resnet50 time: 51.3ms (= 10261.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.5ms (= 13494.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.32 (= 67.5ms / 51.3ms)

OneFlow resnet50 time: 49.1ms (= 9821.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 62.4ms (= 12488.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 62.4ms / 49.1ms)

@mergify mergify bot merged commit 02b1bfd into master Mar 30, 2022
@mergify mergify bot deleted the min_max_observer branch March 30, 2022 17:59
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

@MARD1NO MARD1NO MARD1NO approved these changes

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

Assignees
No one assigned
Labels
automerge eager enhancement
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@liufengwei0103 @hjchen2 @MARD1NO