[PERF] Allreduce fusion. Support torch native matching. Tuning of the thresholds (#24248)

Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Signed-off-by: ilmarkov <markovilya197@gmail.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Robert Shaw <114415538+robertgshaw2-redhat@users.noreply.github.com>
This commit is contained in:
Ilya Markov
2025-11-10 18:33:11 -05:00
committed by GitHub
co-authored by Luka Govedič Luka Govedič Robert Shaw
parent 021143561f
commit d17ecc6b19
6 changed files with 1284 additions and 83 deletions
+7
View File
@@ -71,6 +71,13 @@ if current_platform.is_cuda():
attention_fusions=0,
allreduce_fusions=65,
),
ModelBackendTestCase(
model_name="Qwen/Qwen3-30B-A3B",
model_kwargs=dict(max_model_len=1024),
backend=_Backend.TRITON_ATTN,
attention_fusions=0,
allreduce_fusions=97,
),
]
elif current_platform.is_rocm():