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

Fix partial_fc gradient_funcs bug #8041

Merged
merged 14 commits into from
Apr 20, 2022
Merged

Fix partial_fc gradient_funcs bug #8041

merged 14 commits into from
Apr 20, 2022

Conversation

wyg1997
Copy link
Contributor

@wyg1997 wyg1997 commented Apr 18, 2022

close oneflow-inc/oneteam#1321

@wyg1997 wyg1997 requested a review from doombeaker April 18, 2022 06:30
@wyg1997 wyg1997 changed the title Fix gradient_funcs bug Fix parital_fc gradient_funcs bug Apr 18, 2022
@wyg1997 wyg1997 requested a review from oneflow-ci-bot April 18, 2022 09:07
@github-actions
Copy link
Contributor

Code got formatted by CI. Please request CI again if you still want to have this PR merged. If the PR is from a forked repo, please download the patch files from the GitHub Actions web page and apply them locally.

@marigoold marigoold mentioned this pull request Apr 18, 2022
Closed
@wyg1997 wyg1997 changed the title Fix parital_fc gradient_funcs bug Fix partial_fc gradient_funcs bug Apr 18, 2022
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.6ms (= 12855.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 140.1ms (= 14011.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.09 (= 140.1ms / 128.6ms)

OneFlow resnet50 time: 78.0ms (= 7795.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 81.5ms (= 8146.1ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.04 (= 81.5ms / 78.0ms)

OneFlow resnet50 time: 53.1ms (= 10616.8ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.7ms (= 11540.8ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.09 (= 57.7ms / 53.1ms)

OneFlow resnet50 time: 43.4ms (= 8670.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 44.6ms (= 8912.8ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.03 (= 44.6ms / 43.4ms)

OneFlow resnet50 time: 39.9ms (= 7989.0ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 37.6ms (= 7527.1ms / 200, input_shape=[1, 3, 224, 224])
❌ Relative speed: 0.94 (= 37.6ms / 39.9ms)

OneFlow swin dataloader time: 0.254s (= 50.813s / 200, num_workers=1)
PyTorch swin dataloader time: 0.255s (= 50.925s / 200, num_workers=1)
✔️ Relative speed: 1.002 (= 0.255s / 0.254s)

OneFlow swin dataloader time: 0.069s (= 13.833s / 200, num_workers=4)
PyTorch swin dataloader time: 0.072s (= 14.396s / 200, num_workers=4)
✔️ Relative speed: 1.041 (= 0.072s / 0.069s)

OneFlow swin dataloader time: 0.038s (= 7.502s / 200, num_workers=8)
PyTorch swin dataloader time: 0.037s (= 7.317s / 200, num_workers=8)
✔️ Relative speed: 0.975 (= 0.037s / 0.038s)

✔️ OneFlow resnet50 time: 135.4ms (= 13537.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 157.6ms (= 15763.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.16 (= 157.6ms / 135.4ms)

OneFlow resnet50 time: 88.2ms (= 8816.4ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 106.1ms (= 10606.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 106.1ms / 88.2ms)

OneFlow resnet50 time: 58.6ms (= 11722.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 86.6ms (= 17310.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.48 (= 86.6ms / 58.6ms)

OneFlow resnet50 time: 52.5ms (= 10494.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.2ms (= 13238.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 66.2ms / 52.5ms)

OneFlow resnet50 time: 48.8ms (= 9757.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 61.7ms (= 12332.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 61.7ms / 48.8ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8041/

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.9ms (= 12888.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 139.1ms (= 13912.0ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.08 (= 139.1ms / 128.9ms)

OneFlow resnet50 time: 78.2ms (= 7820.2ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 86.6ms (= 8661.0ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.11 (= 86.6ms / 78.2ms)

OneFlow resnet50 time: 51.8ms (= 10352.1ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.0ms (= 11605.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.12 (= 58.0ms / 51.8ms)

OneFlow resnet50 time: 43.8ms (= 8763.8ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 49.2ms (= 9830.2ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.12 (= 49.2ms / 43.8ms)

OneFlow resnet50 time: 40.1ms (= 8013.2ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.2ms (= 7845.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 0.98 (= 39.2ms / 40.1ms)

OneFlow swin dataloader time: 0.255s (= 51.057s / 200, num_workers=1)
PyTorch swin dataloader time: 0.249s (= 49.814s / 200, num_workers=1)
✔️ Relative speed: 0.976 (= 0.249s / 0.255s)

OneFlow swin dataloader time: 0.067s (= 13.411s / 200, num_workers=4)
PyTorch swin dataloader time: 0.067s (= 13.379s / 200, num_workers=4)
✔️ Relative speed: 0.998 (= 0.067s / 0.067s)

OneFlow swin dataloader time: 0.036s (= 7.239s / 200, num_workers=8)
PyTorch swin dataloader time: 0.036s (= 7.106s / 200, num_workers=8)
✔️ Relative speed: 0.982 (= 0.036s / 0.036s)

✔️ OneFlow resnet50 time: 135.1ms (= 13513.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 155.9ms (= 15588.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 155.9ms / 135.1ms)

OneFlow resnet50 time: 85.9ms (= 8587.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 99.1ms (= 9907.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.15 (= 99.1ms / 85.9ms)

OneFlow resnet50 time: 57.9ms (= 11578.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 74.9ms (= 14981.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.29 (= 74.9ms / 57.9ms)

OneFlow resnet50 time: 52.4ms (= 10470.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.2ms (= 13249.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 66.2ms / 52.4ms)

OneFlow resnet50 time: 50.7ms (= 10142.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 62.8ms (= 12557.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 62.8ms / 50.7ms)

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

✔️ OneFlow resnet50 time: 128.6ms (= 12864.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 139.0ms (= 13901.8ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.08 (= 139.0ms / 128.6ms)

OneFlow resnet50 time: 79.6ms (= 7958.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.1ms (= 8512.8ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.07 (= 85.1ms / 79.6ms)

OneFlow resnet50 time: 52.0ms (= 10392.4ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 60.0ms (= 12000.6ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.15 (= 60.0ms / 52.0ms)

OneFlow resnet50 time: 41.6ms (= 8320.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 48.0ms (= 9605.3ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.15 (= 48.0ms / 41.6ms)

OneFlow resnet50 time: 35.1ms (= 7028.5ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 38.1ms (= 7624.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.08 (= 38.1ms / 35.1ms)

OneFlow swin dataloader time: 0.257s (= 51.321s / 200, num_workers=1)
PyTorch swin dataloader time: 0.250s (= 50.025s / 200, num_workers=1)
✔️ Relative speed: 0.975 (= 0.250s / 0.257s)

OneFlow swin dataloader time: 0.066s (= 13.131s / 200, num_workers=4)
PyTorch swin dataloader time: 0.070s (= 13.943s / 200, num_workers=4)
✔️ Relative speed: 1.062 (= 0.070s / 0.066s)

OneFlow swin dataloader time: 0.036s (= 7.113s / 200, num_workers=8)
PyTorch swin dataloader time: 0.038s (= 7.522s / 200, num_workers=8)
✔️ Relative speed: 1.057 (= 0.038s / 0.036s)

✔️ OneFlow resnet50 time: 135.3ms (= 13528.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 162.8ms (= 16276.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 162.8ms / 135.3ms)

OneFlow resnet50 time: 88.2ms (= 8821.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 105.8ms (= 10579.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 105.8ms / 88.2ms)

OneFlow resnet50 time: 61.6ms (= 12321.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.4ms (= 15072.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 75.4ms / 61.6ms)

OneFlow resnet50 time: 53.5ms (= 10691.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 75.9ms (= 15180.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.42 (= 75.9ms / 53.5ms)

OneFlow resnet50 time: 49.3ms (= 9855.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 61.7ms (= 12346.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 61.7ms / 49.3ms)

@wyg1997 wyg1997 removed the request for review from oneflow-ci-bot April 19, 2022 10:23
@wyg1997 wyg1997 requested a review from oneflow-ci-bot April 19, 2022 10:23
@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/8041/

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 129.6ms (= 12959.4ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 148.8ms (= 14880.5ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.15 (= 148.8ms / 129.6ms)

OneFlow resnet50 time: 82.8ms (= 8282.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 89.8ms (= 8984.6ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.08 (= 89.8ms / 82.8ms)

OneFlow resnet50 time: 57.7ms (= 11533.9ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 59.9ms (= 11989.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.04 (= 59.9ms / 57.7ms)

OneFlow resnet50 time: 46.5ms (= 9299.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.4ms (= 8479.7ms / 200, input_shape=[2, 3, 224, 224])
❌ Relative speed: 0.91 (= 42.4ms / 46.5ms)

OneFlow resnet50 time: 41.5ms (= 8300.1ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.0ms (= 7796.6ms / 200, input_shape=[1, 3, 224, 224])
❌ Relative speed: 0.94 (= 39.0ms / 41.5ms)

OneFlow swin dataloader time: 0.264s (= 52.871s / 200, num_workers=1)
PyTorch swin dataloader time: 0.153s (= 30.678s / 200, num_workers=1)
Relative speed: 0.580 (= 0.153s / 0.264s)

OneFlow swin dataloader time: 0.067s (= 13.445s / 200, num_workers=4)
PyTorch swin dataloader time: 0.043s (= 8.661s / 200, num_workers=4)
Relative speed: 0.644 (= 0.043s / 0.067s)

OneFlow swin dataloader time: 0.040s (= 7.914s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.435s / 200, num_workers=8)
Relative speed: 0.560 (= 0.022s / 0.040s)

❌ OneFlow resnet50 time: 136.8ms (= 13675.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 164.1ms (= 16414.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.20 (= 164.1ms / 136.8ms)

OneFlow resnet50 time: 94.2ms (= 9421.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 103.1ms (= 10313.8ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.09 (= 103.1ms / 94.2ms)

OneFlow resnet50 time: 65.6ms (= 13116.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 80.1ms (= 16015.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 80.1ms / 65.6ms)

OneFlow resnet50 time: 54.8ms (= 10963.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 73.2ms (= 14642.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 73.2ms / 54.8ms)

OneFlow resnet50 time: 54.4ms (= 10878.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.1ms (= 13228.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.22 (= 66.1ms / 54.4ms)

@mergify mergify bot merged commit a89c40e into master Apr 20, 2022
@mergify mergify bot deleted the fix-parital_fc_gradient_funcs branch April 20, 2022 12:22
@BBuf BBuf mentioned this pull request Apr 22, 2022
Closed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@doombeaker doombeaker doombeaker approved these changes

@BBuf BBuf BBuf approved these changes

@daquexian daquexian Awaiting requested review from daquexian

@jackalcooper jackalcooper Awaiting requested review from jackalcooper

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

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

Successfully merging this pull request may close these issues.
4 participants
@wyg1997 @doombeaker @BBuf @oneflow-ci-bot