Update app.py

app.py

@@ -1,5 +1,6 @@
 import os
 import sys
+import subprocess
 from threading import Thread
 from typing import Iterable
 from huggingface_hub import snapshot_download
@@ -19,6 +20,14 @@ from gradio.themes.utils import colors, fonts, sizes
 
 # --- Theme and CSS Definition ---
 
+# Attempt to install flash-attn
+try:
+    subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, check=True, shell=True)
+except subprocess.CalledProcessError as e:
+    print(f"Error installing flash-attn: {e}")
+    print("Continuing without flash-attn.")
+
+
 colors.steel_blue = colors.Color(
     name="steel_blue",
     c50="#EBF3F8",
@@ -153,7 +162,7 @@ MODEL_PATH_D = model_path_d_local
 processor_d = AutoProcessor.from_pretrained(MODEL_PATH_D, trust_remote_code=True)
 model_d = AutoModelForCausalLM.from_pretrained(
     MODEL_PATH_D,
-    attn_implementation="eager",
+    attn_implementation="flash_attention_2",
     torch_dtype=torch.bfloat16,
     device_map="auto",
     trust_remote_code=True