Update vision_analyzer.py

vision_analyzer.py

@@ -76,41 +76,23 @@ import torch
 import subprocess
 import sys
 
-def configure_environment():
-    # Determine device first
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-    
-    # Only attempt flash-attn on CUDA
-    if device == "cuda":
-        try:
-            subprocess.run(
-                [
-                    sys.executable, "-m", "pip", "install",
-                    "flash-attn==2.5.8",  # Pinned version
-                    "--no-build-isolation",
-                    "--no-cache-dir"
-                ],
-                env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "1"},
-                check=True,
-                capture_output=True,
-                text=True
-            )
-            print("Successfully installed flash-attn with CUDA support")
-        except subprocess.CalledProcessError as e:
-            print(f"Flash-attn installation failed (non-critical): {e.stderr}")
-            print("Falling back to standard attention mechanisms")
-    
-    return device
-
-# Usage
-device = configure_environment()
-print(f"Using device: {device}")
-
-
-model_id = "microsoft/Florence-2-large"
-# Either use CPU:
-model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True).eval()
-processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)
+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.")
+
+# Determine the device to use
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+
+try:
+    vision_language_model_large = AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to(device).eval()
+    vision_language_processor_large = AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True)
+except Exception as e:
+    print(f"Error loading large model: {e}")
+    vision_language_model_large = None
+    vision_language_processor_large = None
 
 def log_message(message: str):
     """Log messages with timestamp"""