Update app.py

app.py

@@ -11,16 +11,6 @@ subprocess.run(
     check=False,
 )
 
-from flash_attn import flash_attn_func
-
-try:
-    from ltx_core.model.transformer import attention as _attn_mod
-    _attn_mod.memory_efficient_attention = flash_attn_func
-except Exception as e:
-    print(f"[ATTN] FA3 import failed: {e}, using xformers")
-    from xformers.ops import memory_efficient_attention as _mea
-    _attn_mod.memory_efficient_attention = _mea
-
 # Clone LTX-2 repo and install packages
 LTX_REPO_URL = "https://github.com/Lightricks/LTX-2.git"
 LTX_REPO_DIR = os.path.join(os.path.dirname(os.path.abspath(__file__)), "LTX-2")
@@ -93,6 +83,16 @@ from ltx_pipelines.utils.media_io import decode_audio_from_file, encode_video
 from ltx_core.loader.primitives import LoraPathStrengthAndSDOps
 from ltx_core.loader.sd_ops import LTXV_LORA_COMFY_RENAMING_MAP
 
+from flash_attn import flash_attn_func
+
+try:
+    from ltx_core.model.transformer import attention as _attn_mod
+    _attn_mod.memory_efficient_attention = flash_attn_func
+except Exception as e:
+    print(f"[ATTN] FA3 import failed: {e}, using xformers")
+    from xformers.ops import memory_efficient_attention as _mea
+    _attn_mod.memory_efficient_attention = _mea
+
 # Attention kernels: try FlashAttention-3 first, fall back to xformers
 from ltx_core.model.transformer import attention as _attn_mod
 print(f"[ATTN] Before patch: memory_efficient_attention={_attn_mod.memory_efficient_attention}")