update

app.py

@@ -1,3 +1,7 @@
+
+import subprocess
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+
 import os
 import argparse
 import time
@@ -16,6 +20,7 @@ from huggingface_hub import snapshot_download, hf_hub_download
 # -----------------------------
 # Globals (loaded once per process)
 # -----------------------------
+
 _PIPELINE: Optional[torch.nn.Module] = None
 _DEVICE: Optional[torch.device] = None
 

requirements.txt

@@ -39,5 +39,4 @@ flask-socketio
 torchao
 tensorboard
 ninja
-packaging
-https://github.com/Dao-AILab/flash-attention/releases/download/v2.8.3/flash_attn-2.8.3+cu12torch2.5cxx11abiTRUE-cp310-cp310-linux_x86_64.whl
+packaging