Update app.py

app.py

@@ -23,8 +23,19 @@ import requests
 from RealESRGAN import RealESRGAN
 
 
-import subprocess
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+from unittest.mock import patch
+from typing import Union
+from transformers.dynamic_module_utils import get_imports
+
+
+def fixed_get_imports(filename):
+    """Work around for https://huggingface.co/microsoft/phi-1_5/discussions/72."""
+    if not str(filename).endswith("/modeling_florence2.py"):
+        return get_imports(filename)
+    imports = get_imports(filename)
+    imports.remove("flash_attn")
+    return imports
+
 
 def download_file(url, folder_path, filename):
     if not os.path.exists(folder_path):
@@ -106,8 +117,10 @@ DEFAULT_POSITIVE_SUFFIX = "(masterpiece) very_aesthetic hi_res absurd_res supera
 DEFAULT_NEGATIVE_PREFIX = "Score_1 score_2 score _3 text low_res"
 DEFAULT_NEGATIVE_SUFFIX = "Nsfw oversaturated crappy_art low_quality blurry bad_anatomy extra_digits fewer_digits simple_background very_displeasing watermark signature"
 
-# Initialize Florence model
 device = "cuda" if torch.cuda.is_available() else "cpu"
+
+#def load_models():
+    #with patch("transformers.dynamic_module_utils.get_imports", fixed_get_imports):
 florence_model = AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to(device).eval()
 florence_processor = AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True)