Skip to content

Updated test to include CK/AITER V2/V3 test in single backend case #454

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
Micky774 merged 9 commits into from
Feb 27, 2026
Merged

Updated test to include CK/AITER V2/V3 test in single backend case #454

Changes from all commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
d6e6591
Updated test to include CK/AITER V2/V3 test in single backend case
Micky774 Feb 20, 2026
83c715d
Streamlined per PR review
Micky774 Feb 20, 2026
b0b9906
Updated skip condition
Micky774 Feb 20, 2026
7ac6eab
Added IS_HIP_EXTENSION guard, and included CK/AITER comp in transformer
Micky774 Feb 23, 2026
2fd1fbe
Merge branch 'dev' into zain/ck-pytorch-test
Micky774 Feb 26, 2026
a79d715
Streamlined test
Micky774 Feb 26, 2026
97c612a
Reorganized tests
Micky774 Feb 26, 2026
02a913f
Updated test_transformer_layer
Micky774 Feb 27, 2026
b3c8ce2
Updated test_transformer_layer to be upstream compatible
Micky774 Feb 27, 2026
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
93 changes: 53 additions & 40 deletions tests/pytorch/attention/test_attention.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -240,7 +240,11 @@ def test_dot_product_attention(
flash_attn_supported = True

# Skip if only unfused backend is supported
if (len(fused_attn_backends) + flash_attn_supported + unfused_attn_supported) < 2:
# Double-count the CK backend since we want to compare V2/V3 kernels
has_ck_backend = IS_HIP_EXTENSION and FusedAttnBackend["CK"] in fused_attn_backends
if not has_ck_backend and (
len(fused_attn_backends) + flash_attn_supported + unfused_attn_supported
) < 2:
pytest.skip("Less than two backends to compare.")

# UnfusedDotProductAttention backend
Expand Down Expand Up @@ -271,8 +275,8 @@ def test_dot_product_attention(
)
if len(fused_attn_backends) == 2:
os.environ["NVTE_FUSED_ATTN_BACKEND"] = "0"
os.environ["NVTE_FUSED_ATTN_CK"] = "0"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "1"
os.environ["NVTE_FUSED_ATTN_CK"] = "1"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "0"
fused_attn_fwd, fused_attn_bwd = _run_dot_product_attention(
dtype,
config,
Expand All @@ -284,11 +288,24 @@ def test_dot_product_attention(
is_training,
)
os.environ["NVTE_FUSED_ATTN_BACKEND"] = "1"
os.environ["NVTE_FUSED_ATTN_CK"] = "0"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "1"
fused_attn_fwd_1, fused_attn_bwd_1 = _run_dot_product_attention(
dtype,
config,
"FusedAttention",
ckpt_attn,
qkv_layout,
workspace_opt,
pad_between_seqs,
is_training,
)
if has_ck_backend:
os.environ["NVTE_FUSED_ATTN_CK"] = "1"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "0"
os.environ["NVTE_CK_USES_FWD_V3"] = "1"
os.environ["NVTE_CK_USES_BWD_V3"] = "1"
fused_attn_fwd_1, fused_attn_bwd_1 = _run_dot_product_attention(
os.environ["NVTE_CK_USES_FWD_V3"] = "0"
os.environ["NVTE_CK_USES_BWD_V3"] = "0"
fused_attn_fwd_2, fused_attn_bwd_2 = _run_dot_product_attention(
dtype,
config,
"FusedAttention",
Expand All @@ -298,19 +315,6 @@ def test_dot_product_attention(
pad_between_seqs,
is_training,
)
if IS_HIP_EXTENSION:
os.environ["NVTE_CK_USES_FWD_V3"] = "0"
os.environ["NVTE_CK_USES_BWD_V3"] = "0"
fused_attn_fwd_2, fused_attn_bwd_2 = _run_dot_product_attention(
dtype,
config,
"FusedAttention",
ckpt_attn,
qkv_layout,
workspace_opt,
pad_between_seqs,
is_training,
)


# FlashAttention backend
Expand Down Expand Up @@ -347,11 +351,11 @@ def test_dot_product_attention(
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_1, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_1[i], **tols)
if IS_HIP_EXTENSION:
logging.info("[test_dot_product_attention]: fused attn backend 0 vs 2")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_2, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_2[i], **tols)
if has_ck_backend: # Compare CK V2/V3 if both are available
logging.info("[test_dot_product_attention]: CK fused attn V2 vs V3")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_2, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_2[i], **tols)


@pytest.mark.skipif(get_cudnn_version() < (8, 9, 1), reason="cuDNN 8.9.1+ is required.")
Expand Down Expand Up @@ -1259,7 +1263,11 @@ def test_transformer_layer(
flash_attn_supported, fused_attn_supported, unfused_attn_supported = available_backends

# Skip if only unfused backend is supported
if (len(fused_attn_backends) + flash_attn_supported + unfused_attn_supported) < 2:
# Double-count the CK backend since we want to compare V2/V3 kernels
has_ck_backend = IS_HIP_EXTENSION and FusedAttnBackend["CK"] in fused_attn_backends
if not has_ck_backend and (
len(fused_attn_backends) + flash_attn_supported + unfused_attn_supported
) < 2:
pytest.skip("Less than two backends to compare.")
# Skip if qkv_format = thd and "padding" not in attn_mask_type
if qkv_format == "thd" and "padding" not in config.attn_mask_type:
Expand All @@ -1281,7 +1289,7 @@ def test_transformer_layer(

# FusedAttention backend
if fused_attn_supported:
if len(fused_attn_backends) == 1 or not IS_HIP_EXTENSION:
if len(fused_attn_backends) == 1:
fused_attn_fwd, fused_attn_bwd = _run_transformer_layer(
dtype,
config,
Expand All @@ -1293,9 +1301,9 @@ def test_transformer_layer(
RoPE,
is_training,
)
elif len(fused_attn_backends) == 2:
os.environ["NVTE_FUSED_ATTN_CK"] = "0"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "1"
elif IS_HIP_EXTENSION and len(fused_attn_backends) == 2:
os.environ["NVTE_FUSED_ATTN_CK"] = "1"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "0"
fused_attn_fwd, fused_attn_bwd = _run_transformer_layer(
dtype,
config,
Expand All @@ -1307,8 +1315,8 @@ def test_transformer_layer(
RoPE,
is_training,
)
os.environ["NVTE_FUSED_ATTN_CK"] = "1"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "0"
os.environ["NVTE_FUSED_ATTN_CK"] = "0"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "1"
fused_attn_fwd_1, fused_attn_bwd_1 = _run_transformer_layer(
dtype,
config,
Expand All @@ -1321,6 +1329,9 @@ def test_transformer_layer(
is_training,
)

if has_ck_backend:
os.environ["NVTE_FUSED_ATTN_CK"] = "1"
os.environ["NVTE_FUSED_ATTN_AOTRITON"] = "0"
os.environ["NVTE_CK_USES_FWD_V3"] = "0"
os.environ["NVTE_CK_USES_BWD_V3"] = "0"
fused_attn_fwd_2, fused_attn_bwd_2 = _run_transformer_layer(
Expand Down Expand Up @@ -1363,15 +1374,17 @@ def test_transformer_layer(
logging.info("[test_transformer_layer]: fused attn vs flash attn")
torch.testing.assert_close(fused_attn_fwd, flash_attn_fwd, **tols)
torch.testing.assert_close(fused_attn_bwd, flash_attn_bwd, **tols)
if IS_HIP_EXTENSION and fused_attn_supported and len(fused_attn_backends) == 2:
logging.info("[test_transformer_layer]: fused attn backend 0 vs 1")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_1, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_1[i], **tols)
logging.info("[test_transformer_layer]: fused attn backend 0 vs 2")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_2, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_2[i], **tols)
if IS_HIP_EXTENSION and fused_attn_supported:
if len(fused_attn_backends) == 2:
logging.info("[test_transformer_layer]: fused attn backend 0 vs 1")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_1, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_1[i], **tols)
if has_ck_backend:
logging.info("[test_transformer_layer]: CK fused attn V2 vs V3")
torch.testing.assert_close(fused_attn_fwd, fused_attn_fwd_2, **tols)
for i, _ in enumerate(fused_attn_bwd):
torch.testing.assert_close(fused_attn_bwd[i], fused_attn_bwd_2[i], **tols)


@pytest.mark.skipif(get_cudnn_version() < (8, 9, 1), reason="cuDNN 8.9.1+ is required.")
Expand Down