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.

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

flash attention use mirror #10530

Draft
wants to merge 4 commits into
base: master
Choose a base branch
from
Draft

flash attention use mirror #10530

wants to merge 4 commits into from

Conversation

mosout
Copy link
Contributor

@mosout mosout commented May 22, 2024

No description provided.

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.

@mosout mosout requested review from oneflow-ci-bot and removed request for oneflow-ci-bot May 22, 2024 03:21
Copy link
Contributor

@mosout mosout marked this pull request as draft May 22, 2024 04:21
Copy link
Contributor

Speed stats:
GPU Name: NVIDIA GeForce RTX 3080 Ti 

❌ OneFlow resnet50 time: 43.5ms (= 4345.7ms / 100, input_shape=[16, 3, 224, 224])
PyTorch resnet50 time: 57.9ms (= 5792.6ms / 100, input_shape=[16, 3, 224, 224])
✔️ Relative speed: 1.33 (= 57.9ms / 43.5ms)

OneFlow resnet50 time: 26.4ms (= 2640.1ms / 100, input_shape=[8, 3, 224, 224])
PyTorch resnet50 time: 37.7ms (= 3767.7ms / 100, input_shape=[8, 3, 224, 224])
✔️ Relative speed: 1.43 (= 37.7ms / 26.4ms)

OneFlow resnet50 time: 18.3ms (= 3657.7ms / 200, input_shape=[4, 3, 224, 224])
PyTorch resnet50 time: 36.0ms (= 7192.4ms / 200, input_shape=[4, 3, 224, 224])
✔️ Relative speed: 1.97 (= 36.0ms / 18.3ms)

OneFlow resnet50 time: 17.2ms (= 3443.6ms / 200, input_shape=[2, 3, 224, 224])
PyTorch resnet50 time: 33.8ms (= 6764.0ms / 200, input_shape=[2, 3, 224, 224])
✔️ Relative speed: 1.96 (= 33.8ms / 17.2ms)

OneFlow resnet50 time: 16.9ms (= 3380.3ms / 200, input_shape=[1, 3, 224, 224])
PyTorch resnet50 time: 30.3ms (= 6065.2ms / 200, input_shape=[1, 3, 224, 224])
✔️ Relative speed: 1.79 (= 30.3ms / 16.9ms)

OneFlow swin dataloader time: 0.202s (= 40.387s / 200, num_workers=1)
PyTorch swin dataloader time: 0.128s (= 25.618s / 200, num_workers=1)
Relative speed: 0.634 (= 0.128s / 0.202s)

OneFlow swin dataloader time: 0.063s (= 12.527s / 200, num_workers=4)
PyTorch swin dataloader time: 0.033s (= 6.515s / 200, num_workers=4)
Relative speed: 0.520 (= 0.033s / 0.063s)

OneFlow swin dataloader time: 0.032s (= 6.326s / 200, num_workers=8)
PyTorch swin dataloader time: 0.017s (= 3.325s / 200, num_workers=8)
Relative speed: 0.526 (= 0.017s / 0.032s)

❌ OneFlow resnet50 time: 49.3ms (= 4933.0ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 65.6ms (= 6557.6ms / 100, input_shape=[16, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.33 (= 65.6ms / 49.3ms)

OneFlow resnet50 time: 36.4ms (= 3638.0ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 45.7ms (= 4570.1ms / 100, input_shape=[8, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.26 (= 45.7ms / 36.4ms)

OneFlow resnet50 time: 27.9ms (= 5582.0ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 42.5ms (= 8506.2ms / 200, input_shape=[4, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.52 (= 42.5ms / 27.9ms)

OneFlow resnet50 time: 25.1ms (= 5024.9ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 38.4ms (= 7672.2ms / 200, input_shape=[2, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.53 (= 38.4ms / 25.1ms)

OneFlow resnet50 time: 24.8ms (= 4957.8ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
PyTorch resnet50 time: 36.2ms (= 7236.6ms / 200, input_shape=[1, 3, 224, 224], ddp, world size=2)
✔️ Relative speed: 1.46 (= 36.2ms / 24.8ms)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants