Install flash_attn

pre-requirements.txt

@@ -1 +1 @@
-torch
+-U pip

st_app.py

@@ -5,8 +5,14 @@ try:
     from llmlib.runtime import filled_model_registry
 except ImportError:
     import os
+    import subprocess
 
     os.system("pip install -e ./llmlib")
+    subprocess.run(
+        "pip install flash-attn --no-build-isolation",
+        env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
+        shell=True,
+    )
     from llmlib.runtime import filled_model_registry
 
 from llmlib.model_registry import ModelEntry, ModelRegistry