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

autograd.functional.vjp #10356

Merged
merged 17 commits into from
Dec 21, 2023
Merged

autograd.functional.vjp #10356

merged 17 commits into from
Dec 21, 2023

Conversation

marigoold
Copy link
Contributor

@marigoold marigoold commented Nov 22, 2023 •
edited
Loading

文档：
image

@marigoold marigoold marked this pull request as ready for review November 24, 2023 02:53
@github-actions 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 requested review from oneflow-ci-bot and removed request for oneflow-ci-bot November 24, 2023 02:58
@doombeaker doombeaker requested review from levi131 and removed request for doombeaker November 24, 2023 03:09
@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.8ms (= 4384.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.1ms (= 5714.8ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.30 (= 57.1ms / 43.8ms)

OneFlow resnet50 time: 26.4ms (= 2642.7ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.2ms (= 3815.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.44 (= 38.2ms / 26.4ms)

OneFlow resnet50 time: 18.5ms (= 3697.8ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 36.0ms (= 7195.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.95 (= 36.0ms / 18.5ms)

OneFlow resnet50 time: 17.6ms (= 3529.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 32.4ms (= 6480.2ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.84 (= 32.4ms / 17.6ms)

OneFlow resnet50 time: 17.1ms (= 3416.9ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 28.8ms (= 5769.0ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.69 (= 28.8ms / 17.1ms)

OneFlow swin dataloader time: 0.201s (= 40.143s / 200, num_workers=1)
PyTorch swin dataloader time: 0.127s (= 25.485s / 200, num_workers=1)
Relative speed: 0.635 (= 0.127s / 0.201s)

OneFlow swin dataloader time: 0.056s (= 11.219s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.684s / 200, num_workers=4)
Relative speed: 0.596 (= 0.033s / 0.056s)

OneFlow swin dataloader time: 0.031s (= 6.238s / 200, num_workers=8)
PyTorch swin dataloader time: 0.016s (= 3.281s / 200, num_workers=8)
Relative speed: 0.526 (= 0.016s / 0.031s)

❌ OneFlow resnet50 time: 47.9ms (= 4787.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 64.0ms (= 6395.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 64.0ms / 47.9ms)

OneFlow resnet50 time: 31.7ms (= 3174.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.0ms (= 4503.6ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.42 (= 45.0ms / 31.7ms)

OneFlow resnet50 time: 23.6ms (= 4717.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.9ms (= 8172.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.73 (= 40.9ms / 23.6ms)

OneFlow resnet50 time: 21.4ms (= 4274.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 36.8ms (= 7351.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.72 (= 36.8ms / 21.4ms)

OneFlow resnet50 time: 20.6ms (= 4111.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 34.8ms (= 6954.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.69 (= 34.8ms / 20.6ms)

@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.8ms (= 4382.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 63.0ms (= 6299.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.44 (= 63.0ms / 43.8ms)

OneFlow resnet50 time: 26.1ms (= 2609.2ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.3ms (= 3725.8ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.43 (= 37.3ms / 26.1ms)

OneFlow resnet50 time: 18.9ms (= 3773.6ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.8ms (= 7163.2ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.90 (= 35.8ms / 18.9ms)

OneFlow resnet50 time: 18.0ms (= 3609.4ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.3ms (= 6255.7ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.73 (= 31.3ms / 18.0ms)

OneFlow resnet50 time: 17.6ms (= 3514.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 29.1ms (= 5829.4ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.66 (= 29.1ms / 17.6ms)

OneFlow swin dataloader time: 0.201s (= 40.202s / 200, num_workers=1)
PyTorch swin dataloader time: 0.129s (= 25.777s / 200, num_workers=1)
Relative speed: 0.641 (= 0.129s / 0.201s)

OneFlow swin dataloader time: 0.054s (= 10.818s / 200, num_workers=4)
PyTorch swin dataloader time: 0.032s (= 6.453s / 200, num_workers=4)
Relative speed: 0.597 (= 0.032s / 0.054s)

OneFlow swin dataloader time: 0.032s (= 6.316s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.380s / 200, num_workers=8)
Relative speed: 0.535 (= 0.017s / 0.032s)

❌ OneFlow resnet50 time: 47.9ms (= 4789.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 63.1ms (= 6309.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.32 (= 63.1ms / 47.9ms)

OneFlow resnet50 time: 32.2ms (= 3218.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 44.4ms (= 4443.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 44.4ms / 32.2ms)

OneFlow resnet50 time: 23.9ms (= 4780.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 41.5ms (= 8298.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.74 (= 41.5ms / 23.9ms)

OneFlow resnet50 time: 20.8ms (= 4160.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 36.8ms (= 7358.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.77 (= 36.8ms / 20.8ms)

OneFlow resnet50 time: 20.5ms (= 4100.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 34.5ms (= 6899.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.68 (= 34.5ms / 20.5ms)

@github-actions GitHub Actions
Copy link
Contributor

Speed stats:

@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Dec 4, 2023

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@github-actions GitHub Actions
Copy link
Contributor

github-actions bot commented Dec 4, 2023

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.2ms (= 4321.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 61.3ms (= 6133.0ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.42 (= 61.3ms / 43.2ms)

OneFlow resnet50 time: 26.5ms (= 2645.7ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.4ms (= 3738.8ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.41 (= 37.4ms / 26.5ms)

OneFlow resnet50 time: 17.6ms (= 3514.1ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 34.2ms (= 6832.9ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.94 (= 34.2ms / 17.6ms)

OneFlow resnet50 time: 16.1ms (= 3218.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 33.7ms (= 6746.5ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 2.10 (= 33.7ms / 16.1ms)

OneFlow resnet50 time: 17.1ms (= 3429.4ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 31.0ms (= 6194.2ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.81 (= 31.0ms / 17.1ms)

OneFlow swin dataloader time: 0.214s (= 42.820s / 200, num_workers=1)
PyTorch swin dataloader time: 0.129s (= 25.811s / 200, num_workers=1)
Relative speed: 0.603 (= 0.129s / 0.214s)

OneFlow swin dataloader time: 0.056s (= 11.138s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.562s / 200, num_workers=4)
Relative speed: 0.589 (= 0.033s / 0.056s)

OneFlow swin dataloader time: 0.031s (= 6.246s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.338s / 200, num_workers=8)
Relative speed: 0.534 (= 0.017s / 0.031s)

❌ OneFlow resnet50 time: 49.1ms (= 4907.5ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.4ms (= 6542.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 65.4ms / 49.1ms)

OneFlow resnet50 time: 36.4ms (= 3644.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 49.5ms (= 4952.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.36 (= 49.5ms / 36.4ms)

OneFlow resnet50 time: 30.1ms (= 6027.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 41.4ms (= 8284.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.37 (= 41.4ms / 30.1ms)

OneFlow resnet50 time: 27.3ms (= 5456.7ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.5ms (= 8092.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.48 (= 40.5ms / 27.3ms)

OneFlow resnet50 time: 24.8ms (= 4965.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 35.8ms (= 7151.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.44 (= 35.8ms / 24.8ms)

@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.2ms (= 4316.0ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 61.1ms (= 6111.3ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.42 (= 61.1ms / 43.2ms)

OneFlow resnet50 time: 26.3ms (= 2634.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.3ms (= 3728.0ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.41 (= 37.3ms / 26.3ms)

OneFlow resnet50 time: 17.7ms (= 3546.7ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.9ms (= 7177.0ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 2.02 (= 35.9ms / 17.7ms)

OneFlow resnet50 time: 15.3ms (= 3057.5ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.7ms (= 6347.1ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 2.08 (= 31.7ms / 15.3ms)

OneFlow resnet50 time: 15.9ms (= 3175.4ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 29.9ms (= 5981.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.88 (= 29.9ms / 15.9ms)

OneFlow swin dataloader time: 0.210s (= 41.942s / 200, num_workers=1)
PyTorch swin dataloader time: 0.129s (= 25.721s / 200, num_workers=1)
Relative speed: 0.613 (= 0.129s / 0.210s)

OneFlow swin dataloader time: 0.060s (= 11.925s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.606s / 200, num_workers=4)
Relative speed: 0.554 (= 0.033s / 0.060s)

OneFlow swin dataloader time: 0.030s (= 6.071s / 200, num_workers=8)
PyTorch swin dataloader time: 0.016s (= 3.283s / 200, num_workers=8)
Relative speed: 0.541 (= 0.016s / 0.030s)

❌ OneFlow resnet50 time: 49.2ms (= 4916.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.4ms (= 6844.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.39 (= 68.4ms / 49.2ms)

OneFlow resnet50 time: 37.4ms (= 3737.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 46.3ms (= 4630.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.24 (= 46.3ms / 37.4ms)

OneFlow resnet50 time: 29.0ms (= 5805.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 42.2ms (= 8437.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.45 (= 42.2ms / 29.0ms)

OneFlow resnet50 time: 25.0ms (= 5005.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.8ms (= 8158.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.63 (= 40.8ms / 25.0ms)

OneFlow resnet50 time: 25.9ms (= 5172.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 37.9ms (= 7573.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.46 (= 37.9ms / 25.9ms)

@marigoold marigoold removed the request for review from jackalcooper December 16, 2023 01:04
@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

1 similar comment
@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.5ms (= 4348.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 58.5ms (= 5852.5ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.35 (= 58.5ms / 43.5ms)

OneFlow resnet50 time: 26.6ms (= 2655.8ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.6ms (= 3856.1ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.45 (= 38.6ms / 26.6ms)

OneFlow resnet50 time: 18.3ms (= 3667.1ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.9ms (= 7180.0ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.96 (= 35.9ms / 18.3ms)

OneFlow resnet50 time: 17.8ms (= 3561.7ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.6ms (= 6328.0ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.78 (= 31.6ms / 17.8ms)

OneFlow resnet50 time: 16.6ms (= 3328.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 28.6ms (= 5723.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.72 (= 28.6ms / 16.6ms)

OneFlow swin dataloader time: 0.207s (= 41.431s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.645s / 200, num_workers=1)
Relative speed: 0.619 (= 0.128s / 0.207s)

OneFlow swin dataloader time: 0.059s (= 11.713s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.604s / 200, num_workers=4)
Relative speed: 0.564 (= 0.033s / 0.059s)

OneFlow swin dataloader time: 0.030s (= 5.936s / 200, num_workers=8)
PyTorch swin dataloader time: 0.016s (= 3.287s / 200, num_workers=8)
Relative speed: 0.554 (= 0.016s / 0.030s)

❌ OneFlow resnet50 time: 49.2ms (= 4919.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 66.8ms (= 6683.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.36 (= 66.8ms / 49.2ms)

OneFlow resnet50 time: 36.6ms (= 3661.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 47.8ms (= 4780.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 47.8ms / 36.6ms)

OneFlow resnet50 time: 28.0ms (= 5608.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.8ms (= 8154.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.45 (= 40.8ms / 28.0ms)

OneFlow resnet50 time: 26.2ms (= 5233.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.4ms (= 8076.5ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.54 (= 40.4ms / 26.2ms)

OneFlow resnet50 time: 24.7ms (= 4934.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 36.3ms (= 7267.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.47 (= 36.3ms / 24.7ms)

@marigoold marigoold requested review from oneflow-ci-bot and removed request for oneflow-ci-bot December 19, 2023 03:25
@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.8ms (= 4383.5ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.5ms (= 5749.0ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.31 (= 57.5ms / 43.8ms)

OneFlow resnet50 time: 26.1ms (= 2606.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.4ms (= 3737.2ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.43 (= 37.4ms / 26.1ms)

OneFlow resnet50 time: 18.8ms (= 3760.9ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 34.3ms (= 6851.9ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.82 (= 34.3ms / 18.8ms)

OneFlow resnet50 time: 17.3ms (= 3454.2ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 30.9ms (= 6184.8ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.79 (= 30.9ms / 17.3ms)

OneFlow resnet50 time: 16.7ms (= 3339.2ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 30.5ms (= 6105.2ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.83 (= 30.5ms / 16.7ms)

OneFlow swin dataloader time: 0.200s (= 40.026s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.570s / 200, num_workers=1)
Relative speed: 0.639 (= 0.128s / 0.200s)

OneFlow swin dataloader time: 0.056s (= 11.185s / 200, num_workers=4)
PyTorch swin dataloader time: 0.032s (= 6.469s / 200, num_workers=4)
Relative speed: 0.578 (= 0.032s / 0.056s)

OneFlow swin dataloader time: 0.030s (= 6.053s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.339s / 200, num_workers=8)
Relative speed: 0.552 (= 0.017s / 0.030s)

❌ OneFlow resnet50 time: 48.9ms (= 4893.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.4ms (= 6543.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 65.4ms / 48.9ms)

OneFlow resnet50 time: 36.5ms (= 3645.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.5ms (= 4550.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.25 (= 45.5ms / 36.5ms)

OneFlow resnet50 time: 28.2ms (= 5647.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.3ms (= 8063.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.43 (= 40.3ms / 28.2ms)

OneFlow resnet50 time: 25.0ms (= 4990.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 41.0ms (= 8193.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.64 (= 41.0ms / 25.0ms)

OneFlow resnet50 time: 23.5ms (= 4699.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 37.5ms (= 7490.5ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.59 (= 37.5ms / 23.5ms)

@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.3ms (= 4328.8ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.5ms (= 5749.1ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.33 (= 57.5ms / 43.3ms)

OneFlow resnet50 time: 26.0ms (= 2598.4ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.2ms (= 3817.9ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.47 (= 38.2ms / 26.0ms)

OneFlow resnet50 time: 20.3ms (= 4052.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 34.8ms (= 6953.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.72 (= 34.8ms / 20.3ms)

OneFlow resnet50 time: 17.5ms (= 3498.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.1ms (= 6228.4ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.78 (= 31.1ms / 17.5ms)

OneFlow resnet50 time: 17.2ms (= 3444.2ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 28.5ms (= 5695.9ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.65 (= 28.5ms / 17.2ms)

OneFlow swin dataloader time: 0.199s (= 39.822s / 200, num_workers=1)
PyTorch swin dataloader time: 0.127s (= 25.457s / 200, num_workers=1)
Relative speed: 0.639 (= 0.127s / 0.199s)

OneFlow swin dataloader time: 0.054s (= 10.839s / 200, num_workers=4)
PyTorch swin dataloader time: 0.032s (= 6.486s / 200, num_workers=4)
Relative speed: 0.598 (= 0.032s / 0.054s)

OneFlow swin dataloader time: 0.031s (= 6.159s / 200, num_workers=8)
PyTorch swin dataloader time: 0.016s (= 3.299s / 200, num_workers=8)
Relative speed: 0.536 (= 0.016s / 0.031s)

❌ OneFlow resnet50 time: 49.0ms (= 4904.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.8ms (= 6575.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.34 (= 65.8ms / 49.0ms)

OneFlow resnet50 time: 36.3ms (= 3634.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.9ms (= 4592.5ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 45.9ms / 36.3ms)

OneFlow resnet50 time: 28.1ms (= 5620.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 40.5ms (= 8098.6ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.44 (= 40.5ms / 28.1ms)

OneFlow resnet50 time: 25.1ms (= 5024.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.9ms (= 7779.1ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.55 (= 38.9ms / 25.1ms)

OneFlow resnet50 time: 23.6ms (= 4727.2ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 35.9ms (= 7180.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.52 (= 35.9ms / 23.6ms)

@marigoold marigoold requested review from oneflow-ci-bot and removed request for oneflow-ci-bot December 19, 2023 06:25
@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.6ms (= 4364.8ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.4ms (= 5744.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.32 (= 57.4ms / 43.6ms)

OneFlow resnet50 time: 26.1ms (= 2608.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.3ms (= 3734.5ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.43 (= 37.3ms / 26.1ms)

OneFlow resnet50 time: 18.1ms (= 3613.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.4ms (= 7088.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.96 (= 35.4ms / 18.1ms)

OneFlow resnet50 time: 18.1ms (= 3614.8ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 31.3ms (= 6259.5ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.73 (= 31.3ms / 18.1ms)

OneFlow resnet50 time: 17.4ms (= 3482.0ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 29.2ms (= 5832.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.68 (= 29.2ms / 17.4ms)

OneFlow swin dataloader time: 0.202s (= 40.468s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.504s / 200, num_workers=1)
Relative speed: 0.630 (= 0.128s / 0.202s)

OneFlow swin dataloader time: 0.053s (= 10.594s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.508s / 200, num_workers=4)
Relative speed: 0.614 (= 0.033s / 0.053s)

OneFlow swin dataloader time: 0.031s (= 6.171s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.313s / 200, num_workers=8)
Relative speed: 0.537 (= 0.017s / 0.031s)

❌ OneFlow resnet50 time: 49.1ms (= 4908.9ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.4ms (= 6544.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 65.4ms / 49.1ms)

OneFlow resnet50 time: 36.3ms (= 3626.2ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 46.1ms (= 4611.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.27 (= 46.1ms / 36.3ms)

OneFlow resnet50 time: 28.1ms (= 5629.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 41.0ms (= 8207.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.46 (= 41.0ms / 28.1ms)

OneFlow resnet50 time: 25.3ms (= 5057.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.2ms (= 7640.4ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.51 (= 38.2ms / 25.3ms)

OneFlow resnet50 time: 23.7ms (= 4737.9ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 35.8ms (= 7164.4ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.51 (= 35.8ms / 23.7ms)

levi131
levi131 approved these changes Dec 21, 2023
View reviewed changes
Copy link
Contributor

@levi131 levi131 left a comment

Choose a reason for hiding this comment

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

LGTM

@github-actions GitHub Actions
Copy link
Contributor

View latest API docs preview at: https://oneflow-staging.oss-cn-beijing.aliyuncs.com/docs/Oneflow-Inc/oneflow/pr/10356/

@marigoold marigoold enabled auto-merge (squash) December 21, 2023 07:59
@github-actions GitHub Actions
Copy link
Contributor

Speed stats: 
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.6ms (= 4355.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.4ms (= 5741.8ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.32 (= 57.4ms / 43.6ms)

OneFlow resnet50 time: 26.2ms (= 2622.2ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 38.5ms (= 3848.5ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.47 (= 38.5ms / 26.2ms)

OneFlow resnet50 time: 18.3ms (= 3660.6ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 35.6ms (= 7121.5ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.95 (= 35.6ms / 18.3ms)

OneFlow resnet50 time: 17.5ms (= 3490.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 29.9ms (= 5974.5ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.71 (= 29.9ms / 17.5ms)

OneFlow resnet50 time: 17.0ms (= 3393.5ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 29.9ms (= 5972.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.76 (= 29.9ms / 17.0ms)

OneFlow swin dataloader time: 0.200s (= 40.091s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.517s / 200, num_workers=1)
Relative speed: 0.636 (= 0.128s / 0.200s)

OneFlow swin dataloader time: 0.054s (= 10.860s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.622s / 200, num_workers=4)
Relative speed: 0.610 (= 0.033s / 0.054s)

OneFlow swin dataloader time: 0.030s (= 6.025s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.318s / 200, num_workers=8)
Relative speed: 0.551 (= 0.017s / 0.030s)

❌ OneFlow resnet50 time: 48.8ms (= 4875.8ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 63.9ms (= 6387.1ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 63.9ms / 48.8ms)

OneFlow resnet50 time: 36.4ms (= 3640.6ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 49.3ms (= 4930.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.35 (= 49.3ms / 36.4ms)

OneFlow resnet50 time: 28.2ms (= 5635.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 39.1ms (= 7829.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.39 (= 39.1ms / 28.2ms)

OneFlow resnet50 time: 24.9ms (= 4982.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.3ms (= 7654.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.54 (= 38.3ms / 24.9ms)

OneFlow resnet50 time: 24.0ms (= 4790.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 37.1ms (= 7427.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.55 (= 37.1ms / 24.0ms)

@marigoold marigoold merged commit 559f5ec into master Dec 21, 2023
20 checks passed
@marigoold marigoold deleted the dev_add_autograd_functional_vjp branch December 21, 2023 08:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@akeeei akeeei akeeei approved these changes

@levi131 levi131 levi131 approved these changes

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

Assignees
No one assigned
Labels
api documentation enhancement need-clean-ccache python
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@marigoold @levi131 @akeeei @oneflow-ci-bot