Doing the subprocess install before importation

app.py

@@ -1,4 +1,10 @@
 import subprocess
+
+subprocess.run(
+    "pip install flash-attn --no-build-isolation",
+    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+    shell=True,
+)
 import uuid
 
 import gradio as gr
@@ -8,12 +14,6 @@ from qwen_vl_utils import process_vision_info
 from transformers import AutoProcessor, Qwen2VLForConditionalGeneration
 from voyager_index import Voyager
 
-subprocess.run(
-    "pip install flash-attn --no-build-isolation",
-    env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
-    shell=True,
-)
-
 device = "cuda" if torch.cuda.is_available() else "cpu"