Skip to content

Fix numerical stability of test_gemm_common.py #283

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

Merged
merged 2 commits into from
Sep 5, 2025
Merged

Fix numerical stability of test_gemm_common.py #283

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e500c75
Fix numerical stability of test_gemm_common.py
cjluo-nv Sep 4, 2025
29c2435
Merge branch 'main' into chenjiel/cjluo-nv-patch-1
cjluo-nv Sep 4, 2025
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
14 changes: 10 additions & 4 deletions tests/gpu/torch/quantization/backends/test_gemm_common.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,12 @@
set_seed()


@pytest.fixture(autouse=True)
def setup_seed():
"""Set seed before each test function."""
set_seed()


@pytest.mark.parametrize(
("config", "gemm_forward", "atol", "rtol"),
[
Expand Down Expand Up @@ -257,9 +263,9 @@ def forward_loop(model, run_backward=False):

# The way the compression of the weights and inputs might be different.
# E.g. we may use torch.compile in the gemms.
assert torch.allclose(output_dynamic_quant_gemm, output_dynamic_quant, atol=atol / 3)
assert torch.allclose(output_calib_quant_gemm, output_calib_quant, atol=atol / 3)
assert torch.allclose(output_dynamic_quant_gemm, output_dynamic_quant, atol=atol / 2)
assert torch.allclose(output_calib_quant_gemm, output_calib_quant, atol=atol / 2)
assert torch.allclose(
output_dynamic_quant_gemm, output_dynamic_quant_compressed, atol=atol / 3
output_dynamic_quant_gemm, output_dynamic_quant_compressed, atol=atol / 2
)
assert torch.allclose(output_calib_quant_gemm, output_calib_quant_compressed, atol=atol / 3)
assert torch.allclose(output_calib_quant_gemm, output_calib_quant_compressed, atol=atol / 2)