Skip to content

Honor the NVTE_FUSED_ATTN_<backend> in test_fused_attn.py #123

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 1 commit into
base: dev
Choose a base branch
from
+11 −2
Open

Honor the NVTE_FUSED_ATTN_<backend> in test_fused_attn.py #123

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
13 changes: 11 additions & 2 deletions tests/pytorch/fused_attn/test_fused_attn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -195,17 +195,26 @@ def test():

if IS_HIP_EXTENSION:
backends = {"AOTriton": "AOTRITON", "CK": "CK"}
# record the backend enablement from envs
backend_enabled = {}
for i in backends.keys():
backend_enabled[backends[i]] = os.getenv("NVTE_FUSED_ATTN_"+backends[i], "1")
with logging_context():
for i in backends.keys():
# skip the already disabled backend
if int(backend_enabled[backends[i]])==0:
continue
# disable other backends
for k in backends.keys():
os.environ["NVTE_FUSED_ATTN_"+backends[k]] = "0"
os.environ["NVTE_FUSED_ATTN_"+backends[i]] = "1"
_attention_backends["backend_selection_requires_update"] = True
available_backends, fused_attention_backend = test()
# restore backend enablement envs
for k in backends.keys():
os.environ["NVTE_FUSED_ATTN_"+backends[k]] = backend_enabled[backends[k]]
if fused_attention_backend == FusedAttnBackend[i]:
fused_attn_backends.append(fused_attention_backend)
for i in backends.keys():
del os.environ["NVTE_FUSED_ATTN_"+backends[i]];
else:
backends = {0: "F16_max512_seqlen", 1: "F16_arbitrary_seqlen", 2: "FP8"}
with logging_context():
Expand Down