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 tril_ #9996

Merged
merged 5 commits into from
Mar 18, 2023
Merged

Add tril_ #9996

merged 5 commits into from
Mar 18, 2023

Conversation

liujuncheng
Copy link
Collaborator

No description provided.

@liujuncheng liujuncheng requested a review from hjchen2 as a code owner March 17, 2023 06:55
@liujuncheng liujuncheng enabled auto-merge (squash) March 17, 2023 07:54
@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.9ms (= 14086.0ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.7ms (= 14273.9ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 142.7ms / 140.9ms)

OneFlow resnet50 time: 80.4ms (= 8043.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 81.4ms (= 8137.3ms / 100, input_shape=[8, 3, 224, 224])
❌ Relative speed: 1.01 (= 81.4ms / 80.4ms)

OneFlow resnet50 time: 48.5ms (= 9696.6ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 58.5ms (= 11692.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.21 (= 58.5ms / 48.5ms)

OneFlow resnet50 time: 31.8ms (= 6364.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 45.4ms (= 9080.6ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.43 (= 45.4ms / 31.8ms)

OneFlow resnet50 time: 24.9ms (= 4985.4ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 41.3ms (= 8252.9ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.66 (= 41.3ms / 24.9ms)

OneFlow swin dataloader time: 0.235s (= 46.957s / 200, num_workers=1)
PyTorch swin dataloader time: 0.147s (= 29.427s / 200, num_workers=1)
Relative speed: 0.627 (= 0.147s / 0.235s)

OneFlow swin dataloader time: 0.069s (= 13.733s / 200, num_workers=4)
PyTorch swin dataloader time: 0.043s (= 8.586s / 200, num_workers=4)
Relative speed: 0.625 (= 0.043s / 0.069s)

OneFlow swin dataloader time: 0.041s (= 8.252s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.593s / 200, num_workers=8)
Relative speed: 0.557 (= 0.023s / 0.041s)

❌ OneFlow resnet50 time: 152.0ms (= 15199.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.8ms (= 16076.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.06 (= 160.8ms / 152.0ms)

OneFlow resnet50 time: 90.6ms (= 9058.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 103.1ms (= 10310.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.14 (= 103.1ms / 90.6ms)

OneFlow resnet50 time: 59.1ms (= 11817.8ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 81.4ms (= 16277.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.38 (= 81.4ms / 59.1ms)

OneFlow resnet50 time: 42.2ms (= 8443.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 72.6ms (= 14513.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.72 (= 72.6ms / 42.2ms)

OneFlow resnet50 time: 35.6ms (= 7123.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.0ms (= 13595.1ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.91 (= 68.0ms / 35.6ms)

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.8ms (= 14080.8ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 143.8ms (= 14377.1ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.02 (= 143.8ms / 140.8ms)

OneFlow resnet50 time: 80.4ms (= 8038.9ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.0ms (= 8498.2ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.06 (= 85.0ms / 80.4ms)

OneFlow resnet50 time: 48.4ms (= 9674.9ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 57.5ms (= 11504.4ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.19 (= 57.5ms / 48.4ms)

OneFlow resnet50 time: 32.6ms (= 6515.0ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 46.0ms (= 9191.1ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.41 (= 46.0ms / 32.6ms)

OneFlow resnet50 time: 25.4ms (= 5089.2ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 40.5ms (= 8108.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.59 (= 40.5ms / 25.4ms)

OneFlow swin dataloader time: 0.236s (= 47.299s / 200, num_workers=1)
PyTorch swin dataloader time: 0.150s (= 29.968s / 200, num_workers=1)
Relative speed: 0.634 (= 0.150s / 0.236s)

OneFlow swin dataloader time: 0.068s (= 13.561s / 200, num_workers=4)
PyTorch swin dataloader time: 0.044s (= 8.770s / 200, num_workers=4)
Relative speed: 0.647 (= 0.044s / 0.068s)

OneFlow swin dataloader time: 0.041s (= 8.199s / 200, num_workers=8)
PyTorch swin dataloader time: 0.023s (= 4.617s / 200, num_workers=8)
Relative speed: 0.563 (= 0.023s / 0.041s)

❌ OneFlow resnet50 time: 152.0ms (= 15202.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 161.1ms (= 16107.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.06 (= 161.1ms / 152.0ms)

OneFlow resnet50 time: 90.7ms (= 9073.3ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 101.4ms (= 10138.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 101.4ms / 90.7ms)

OneFlow resnet50 time: 58.9ms (= 11785.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.2ms (= 15646.3ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 78.2ms / 58.9ms)

OneFlow resnet50 time: 42.3ms (= 8455.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 71.9ms (= 14372.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.70 (= 71.9ms / 42.3ms)

OneFlow resnet50 time: 36.7ms (= 7341.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 68.4ms (= 13678.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.86 (= 68.4ms / 36.7ms)

@github-actions
Copy link
Contributor

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

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.9ms (= 14093.6ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 140.4ms (= 14037.4ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.00 (= 140.4ms / 140.9ms)

OneFlow resnet50 time: 80.5ms (= 8053.0ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.4ms (= 8542.3ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.06 (= 85.4ms / 80.5ms)

OneFlow resnet50 time: 49.0ms (= 9791.3ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 63.0ms (= 12606.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.29 (= 63.0ms / 49.0ms)

OneFlow resnet50 time: 32.3ms (= 6457.3ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.7ms (= 8542.7ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.32 (= 42.7ms / 32.3ms)

OneFlow resnet50 time: 25.0ms (= 5003.3ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 34.2ms (= 6845.8ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.37 (= 34.2ms / 25.0ms)

OneFlow swin dataloader time: 0.235s (= 47.009s / 200, num_workers=1)
PyTorch swin dataloader time: 0.148s (= 29.603s / 200, num_workers=1)
Relative speed: 0.630 (= 0.148s / 0.235s)

OneFlow swin dataloader time: 0.069s (= 13.894s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.309s / 200, num_workers=4)
Relative speed: 0.598 (= 0.042s / 0.069s)

OneFlow swin dataloader time: 0.043s (= 8.509s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.455s / 200, num_workers=8)
Relative speed: 0.524 (= 0.022s / 0.043s)

❌ OneFlow resnet50 time: 152.3ms (= 15229.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 159.9ms (= 15985.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.05 (= 159.9ms / 152.3ms)

OneFlow resnet50 time: 90.9ms (= 9089.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 102.1ms (= 10205.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.12 (= 102.1ms / 90.9ms)

OneFlow resnet50 time: 59.7ms (= 11932.4ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.3ms (= 15652.9ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.31 (= 78.3ms / 59.7ms)

OneFlow resnet50 time: 41.9ms (= 8387.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 71.4ms (= 14282.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.70 (= 71.4ms / 41.9ms)

OneFlow resnet50 time: 37.4ms (= 7486.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 69.9ms (= 13984.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.87 (= 69.9ms / 37.4ms)

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 141.0ms (= 14096.4ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 141.7ms (= 14173.9ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 141.7ms / 141.0ms)

OneFlow resnet50 time: 80.7ms (= 8065.4ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 83.4ms (= 8344.4ms / 100, input_shape=[8, 3, 224, 224])
❌ Relative speed: 1.03 (= 83.4ms / 80.7ms)

OneFlow resnet50 time: 49.2ms (= 9836.8ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 56.9ms (= 11384.3ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.16 (= 56.9ms / 49.2ms)

OneFlow resnet50 time: 33.4ms (= 6681.1ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 40.9ms (= 8180.3ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.22 (= 40.9ms / 33.4ms)

OneFlow resnet50 time: 25.1ms (= 5025.7ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 43.4ms (= 8686.6ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.73 (= 43.4ms / 25.1ms)

OneFlow swin dataloader time: 0.240s (= 47.926s / 200, num_workers=1)
PyTorch swin dataloader time: 0.149s (= 29.805s / 200, num_workers=1)
Relative speed: 0.622 (= 0.149s / 0.240s)

OneFlow swin dataloader time: 0.072s (= 14.365s / 200, num_workers=4)
PyTorch swin dataloader time: 0.040s (= 8.056s / 200, num_workers=4)
Relative speed: 0.561 (= 0.040s / 0.072s)

OneFlow swin dataloader time: 0.043s (= 8.504s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.497s / 200, num_workers=8)
Relative speed: 0.529 (= 0.022s / 0.043s)

❌ OneFlow resnet50 time: 152.4ms (= 15237.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.5ms (= 16052.4ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.05 (= 160.5ms / 152.4ms)

OneFlow resnet50 time: 91.1ms (= 9106.7ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 102.5ms (= 10251.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.13 (= 102.5ms / 91.1ms)

OneFlow resnet50 time: 59.4ms (= 11874.7ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.8ms (= 15756.1ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 78.8ms / 59.4ms)

OneFlow resnet50 time: 42.8ms (= 8562.6ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.1ms (= 13417.8ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.57 (= 67.1ms / 42.8ms)

OneFlow resnet50 time: 36.1ms (= 7218.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.1ms (= 14023.3ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.94 (= 70.1ms / 36.1ms)

@github-actions
Copy link
Contributor

Speed stats: 
GPU Name: GeForce GTX 1080 

❌ OneFlow resnet50 time: 140.9ms (= 14094.2ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 142.8ms (= 14281.5ms / 100, input_shape=[16, 3, 224, 224])
❌ Relative speed: 1.01 (= 142.8ms / 140.9ms)

OneFlow resnet50 time: 80.6ms (= 8062.8ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 85.3ms (= 8525.1ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.06 (= 85.3ms / 80.6ms)

OneFlow resnet50 time: 49.4ms (= 9881.2ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 56.2ms (= 11230.8ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.14 (= 56.2ms / 49.4ms)

OneFlow resnet50 time: 32.9ms (= 6577.2ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 42.7ms (= 8533.5ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.30 (= 42.7ms / 32.9ms)

OneFlow resnet50 time: 24.3ms (= 4866.1ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 39.4ms (= 7879.4ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.62 (= 39.4ms / 24.3ms)

OneFlow swin dataloader time: 0.236s (= 47.141s / 200, num_workers=1)
PyTorch swin dataloader time: 0.156s (= 31.169s / 200, num_workers=1)
Relative speed: 0.661 (= 0.156s / 0.236s)

OneFlow swin dataloader time: 0.072s (= 14.343s / 200, num_workers=4)
PyTorch swin dataloader time: 0.042s (= 8.389s / 200, num_workers=4)
Relative speed: 0.585 (= 0.042s / 0.072s)

OneFlow swin dataloader time: 0.041s (= 8.236s / 200, num_workers=8)
PyTorch swin dataloader time: 0.022s (= 4.436s / 200, num_workers=8)
Relative speed: 0.539 (= 0.022s / 0.041s)

❌ OneFlow resnet50 time: 152.0ms (= 15202.3ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 160.8ms (= 16079.2ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
❌ Relative speed: 1.06 (= 160.8ms / 152.0ms)

OneFlow resnet50 time: 91.1ms (= 9106.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 102.5ms (= 10247.9ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.13 (= 102.5ms / 91.1ms)

OneFlow resnet50 time: 59.4ms (= 11881.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 78.3ms (= 15662.5ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.32 (= 78.3ms / 59.4ms)

OneFlow resnet50 time: 41.9ms (= 8371.0ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 70.8ms (= 14154.3ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.69 (= 70.8ms / 41.9ms)

OneFlow resnet50 time: 36.1ms (= 7218.0ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 67.2ms (= 13435.7ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.86 (= 67.2ms / 36.1ms)

@github-actions
Copy link
Contributor

View latest API docs preview at: https://staging.oneflow.info/docs/Oneflow-Inc/oneflow/pr/9996/

@liujuncheng liujuncheng merged commit 52d63d7 into master Mar 18, 2023
@liujuncheng liujuncheng deleted the dev_inplace_tril branch March 18, 2023 05:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@leaves-zwx leaves-zwx leaves-zwx approved these changes

@ofhwei ofhwei ofhwei approved these changes

@hjchen2 hjchen2 Awaiting requested review from hjchen2 hjchen2 is a code owner

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

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

Successfully merging this pull request may close these issues.
3 participants
@liujuncheng @leaves-zwx @ofhwei