Feature(MInference): chaning flash_attn

app.py

@@ -1,3 +1,11 @@
+import subprocess
+# Install flash attention, skipping CUDA build if necessary
+subprocess.run(
+    "pip install flash-attn --no-build-isolation",
+    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+    shell=True,
+)
+
 import gradio as gr
 import os
 import spaces

requirements.txt

@@ -1,4 +1,3 @@
-flash_attn
 triton==2.1.0
 pycuda==2023.1
 accelerate