refactor: separate extension ops and backends

This commit is contained in:
2026-08-16 21:15:52 +08:00
parent 3406157431
commit 6ac3b51496
18 changed files with 82 additions and 54 deletions
+2 -7
View File
@@ -10,6 +10,8 @@ from astrai.extension import (
ATTN_BACKEND,
AttentionBackendFactory,
CudaBackend,
FlashAttnBackend,
TorchNativeBackend,
attn_backend,
get_backend,
)
@@ -17,13 +19,6 @@ from astrai.extension import (
def test_default_backend_resolves_to_available():
"""Default backend is the first available in cuda > flash > torch order."""
from astrai.extension.attention_backend import (
CudaBackend,
FlashAttnBackend,
TorchNativeBackend,
_resolve_default_backend,
)
backend = get_backend()
assert isinstance(backend, (CudaBackend, FlashAttnBackend, TorchNativeBackend))