Skip to content

Adjusted accuracy tolerance for unit tests on MI350/MI300 #2590

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

Open
wants to merge 2 commits into
base: release/2.7
Choose a base branch
from
+6 −1
Open

Adjusted accuracy tolerance for unit tests on MI350/MI300 #2590

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
03a3f9b
Adjusted accuracy tolerance for MI350 on test_transformers
BLOrange-AMD Sep 2, 2025
9d5618f
[ROCm][CI] fix mi300 test failure after 6.4.1 update (#156368)
jeffdaily Jun 19, 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
3 changes: 2 additions & 1 deletion test/test_linalg.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9003,7 +9003,8 @@ def dims_full_for_fn():
r1 = fntorch(t0_full, t1, t2)
self.assertEqual(r0, r1)

@tf32_on_and_off(0.001)
# ROCm 6.4 passes with tf32=on, but 6.4.1 needed tolerance reduced slightly
@tf32_on_and_off(0.002 if torch.version.hip else 0.001)
@bf32_on_and_off(0.001)
def test_broadcast_batched_matmul(self, device):
n_dim = random.randint(1, 8)
Expand Down
4 changes: 4 additions & 0 deletions test/test_transformers.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3194,6 +3194,8 @@ def _get_mem_eff_drop_mask(batch_size, n_heads, q_len, kv_len, p, seed, offset,
fudge_factors['grad_query'] = 650.0
if dtype == torch.float32:
fudge_factors['grad_key'] = 90.0
if "gfx95" in torch.cuda.get_device_properties(0).gcnArchName:
fudge_factors['grad_value'] = 15.0

check_out_and_grad(
(out_ref, out_lp_ref, out),
Expand Down Expand Up @@ -3315,6 +3317,8 @@ def _get_mem_eff_drop_mask(batch_size, n_heads, q_len, kv_len, p, seed, offset,
fudge_factors['grad_query'] = 650.0
if dtype == torch.float32:
fudge_factors['grad_key'] = 90.0
if "gfx95" in torch.cuda.get_device_properties(0).gcnArchName:
fudge_factors['grad_value'] = 15.0

check_out_and_grad(
(out_ref, out_lp_ref, out),
Expand Down