[Attention] Refactor FA block_size limitations to hybrid models only (#29084)

Signed-off-by: NickLucche <nlucches@redhat.com>
This commit is contained in:
Nicolò Lucchesi
2025-11-22 06:38:44 -08:00
committed by GitHub
parent 5f7209a793
commit 066209a045
17 changed files with 82 additions and 32 deletions
+1 -1
View File
@@ -61,7 +61,7 @@ for backend in BACKENDS_TO_TEST:
BACKEND_BLOCK_SIZES = {}
for backend in BACKENDS_TO_TEST:
supported_sizes = backend.get_class().supported_kernel_block_sizes
supported_sizes = backend.get_class().get_supported_kernel_block_sizes()
if supported_sizes:
default_size = supported_sizes[0]
block_size = (
+3 -1
View File
@@ -185,7 +185,9 @@ def _make_mock_backend_for_kernel_block_size(
supported_sizes: list[int | MultipleOf],
):
class _MockBackend:
supported_kernel_block_sizes = supported_sizes
@staticmethod
def get_supported_kernel_block_sizes():
return supported_sizes
return _MockBackend()