Skip to content

Adjust 4bit test tolerance on CPU for larger blocksizes #1749

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
matthewdouglas merged 1 commit into from
Sep 8, 2025
Merged

Adjust 4bit test tolerance on CPU for larger blocksizes #1749

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
8 changes: 6 additions & 2 deletions tests/test_functional.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1169,8 +1169,12 @@ def test_4bit_quant(self, device, dtype, quant_type, blocksize):
4096: 0.262457,
}

assert err < error_dict[quant_type]["err"][blocksize] + 1e-3
assert relerr < error_dict[quant_type]["rel_err"][blocksize] + 1e-3
# Allow higher tolerance for fp32 on CPU with larger block sizes
reltol = 2.8e-3 if dtype == torch.float32 and blocksize >= 128 and device == "cpu" else 1e-3
errtol = 1.2e-3 if dtype == torch.float32 and blocksize >= 1024 and device == "cpu" else 1e-3

assert err < error_dict[quant_type]["err"][blocksize] + errtol
assert relerr < error_dict[quant_type]["rel_err"][blocksize] + reltol

@pytest.mark.parametrize("device", get_available_devices())
@pytest.mark.parametrize("quant_type", ["fp4", "nf4"])
Expand Down
Loading