diff --git a/fastvideo/v1/configs/models/dits/base.py b/fastvideo/v1/configs/models/dits/base.py
index c247ca7ed..e5fd1d829 100644
--- a/fastvideo/v1/configs/models/dits/base.py
+++ b/fastvideo/v1/configs/models/dits/base.py
@@ -12,6 +12,7 @@ class DiTArchConfig(ArchConfig):
     _param_names_mapping: dict = field(default_factory=dict)
     _supported_attention_backends: Tuple[_Backend,
                                          ...] = (_Backend.SLIDING_TILE_ATTN,
+                                                 _Backend.SAGE_ATTN,
                                                  _Backend.FLASH_ATTN,
                                                  _Backend.TORCH_SDPA)
 
diff --git a/fastvideo/v1/models/dits/base.py b/fastvideo/v1/models/dits/base.py
index 477c71e4d..3f405d426 100644
--- a/fastvideo/v1/models/dits/base.py
+++ b/fastvideo/v1/models/dits/base.py
@@ -17,8 +17,8 @@ class BaseDiT(nn.Module, ABC):
     num_attention_heads: int
     num_channels_latents: int
     # always supports torch_sdpa
-    _supported_attention_backends: Tuple[_Backend,
-                                         ...] = (_Backend.TORCH_SDPA, )
+    _supported_attention_backends: Tuple[
+        _Backend, ...] = DiTConfig()._supported_attention_backends
 
     def __init_subclass__(cls) -> None:
         required_class_attrs = [