Update app.py

app.py

@@ -1,6 +1,8 @@
+import subprocess
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+
 import os
 import sys
-import subprocess
 import torch
 import datetime
 import numpy as np
@@ -142,23 +144,6 @@ try:
     from hyvideo.commons.infer_state import initialize_infer_state
     # Import the specific I2V System Prompt from the repo
     from hyvideo.utils.rewrite.i2v_prompt import i2v_rewrite_system_prompt
-    
-    # --- FIX: Force Disable Flash Attention Patch ---
-    import hyvideo.models.transformers.modules.attention
-    
-    print("🛠️  Patching Attention Mode to 'torch' (SDPA) to bypass Flash Attn check...")
-    
-    def patched_fallback(attn_mode, infer_state=None, block_idx=None):
-        # Always return 'torch' to bypass the flash-attn check
-        return "torch"
-    
-    # Patch the source definition in commons
-    hyvideo.commons.maybe_fallback_attn_mode = patched_fallback
-    
-    # Patch the reference inside the attention module (crucial for TokenRefiner which imports it)
-    hyvideo.models.transformers.modules.attention.maybe_fallback_attn_mode = patched_fallback
-    # ------------------------------------------------
-
 except ImportError as e:
     print(f"CRITICAL ERROR: {e}")
     sys.exit(1)