Skip to content

Temporary Fix: Skip TestAffineQuantizedTensorParallel on H100 #1001

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

Closed
wants to merge 5 commits into from
Closed

Temporary Fix: Skip TestAffineQuantizedTensorParallel on H100 #1001

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
2b83972
Temporary Fix: Skip on H100
jainapurva Oct 3, 2024
6bf6caa
Temporary Fix: Skip on H100
jainapurva Oct 3, 2024
ce5273c
Temporary Fix: Skip on H100
jainapurva Oct 3, 2024
59a4113
Test cuda nightly failure
jainapurva Oct 3, 2024
9dec2af
Test cuda nightly failure
jainapurva Oct 3, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 8 additions & 3 deletions .github/workflows/regression_test.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,12 +38,17 @@ jobs:
torch-spec: 'torch==2.4.0'
gpu-arch-type: "cuda"
gpu-arch-version: "12.1"
- name: CUDA Nightly
- name: CUDA Nightly (Oct 1)
runs-on: linux.g5.12xlarge.nvidia.gpu
torch-spec: '--pre torch --index-url https://download.pytorch.org/whl/nightly/cu121'
torch-spec: '--pre torch==2.6.0.dev20241001+cu121 --index-url https://download.pytorch.org/whl/nightly/cu121' #'--pre torch --index-url https://download.pytorch.org/whl/nightly/cu121'
gpu-arch-type: "cuda"
gpu-arch-version: "12.1"

- name: CUDA Nightly (Aug 30)
runs-on: linux.g5.12xlarge.nvidia.gpu
torch-spec: '--pre torch==2.5.0.dev20240831+cu121 --index-url https://download.pytorch.org/whl/nightly/cu121' #'--pre torch --index-url https://download.pytorch.org/whl/nightly/cu121'
gpu-arch-type: "cuda"
gpu-arch-version: "12.1"

- name: CPU 2.2.2
runs-on: linux.4xlarge
torch-spec: 'torch==2.2.2 --index-url https://download.pytorch.org/whl/cpu "numpy<2" '
Expand Down
7 changes: 6 additions & 1 deletion test/dtypes/test_affine_quantized_tensor_parallel.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,12 +1,17 @@
from torchao.testing.utils import copy_tests, TorchAOTensorParallelTestCase
from torch.testing._internal.common_utils import run_tests
from torchao.quantization import int8_weight_only
import torch
Copy link
Member

Choose a reason for hiding this comment

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

do you still see the error if you move the import torch statement above any ao imports?

Copy link
Contributor Author

Choose a reason for hiding this comment

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

Moving torch as the first import does resolve some issues, but there are still issues of compatibility with H100. I'm working on another PR to add support for H100 (specially float8). #1003


class TestAffineQuantizedTensorParallel(TorchAOTensorParallelTestCase):
pass

is_H100 = torch.cuda.is_available() and torch.cuda.get_device_capability() >= (9, 0)

copy_tests(TorchAOTensorParallelTestCase, TestAffineQuantizedTensorParallel, "aqt_tp")

if __name__ == "__main__":
run_tests()
if not is_H100:
run_tests()
else:
print("Skipping TestAffineQuantizedTensorParallel: not supported on H100")
Copy link
Member

Choose a reason for hiding this comment

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

I'd prefer we explicitly put a skip test on the skipped test, this will also mess up the pytest output

Loading