Update app.py

app.py

@@ -22,6 +22,20 @@ from transformers import AutoProcessor, AutoModelForCausalLM, pipeline
 import requests
 from RealESRGAN import RealESRGAN
 
+import os
+
+from typing import Union
+from transformers.dynamic_module_utils import get_imports
+
+
+def fixed_get_imports(filename: Union[str, os.PathLike]) -> list[str]:
+    """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
+
 
 import subprocess
 #subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
@@ -108,8 +122,14 @@ DEFAULT_NEGATIVE_SUFFIX = "Nsfw oversaturated crappy_art low_quality blurry bad_
 
 # Initialize Florence model
 device = "cuda" if torch.cuda.is_available() else "cpu"
-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)
+
+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)
+    return florence_model, florence_processor
+
+florence_model, florence_processor = load_models()
 
 # Prompt Enhancer
 enhancer_medium = pipeline("summarization", model="gokaygokay/Lamini-Prompt-Enchance", device=device)