Update model_loader.py

model_loader.py

@@ -1,22 +1,54 @@
+# import torch
+# from transformers import AutoProcessor, AutoModelForVision2Seq
+
+# MODEL_NAME = "Qwen/Qwen2.5-VL-7B-Instruct"
+
+# device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# print("Loading processor...")
+# processor = AutoProcessor.from_pretrained(
+#     MODEL_NAME, 
+#     trust_remote_code=True,
+#     use_fast=True)  # use_fast to avoid warnings in logs
+
+# print("Loading model...")
+# model = AutoModelForVision2Seq.from_pretrained(
+#     MODEL_NAME,
+#     trust_remote_code=True,
+#     torch_dtype=torch.float16,
+#     device_map="auto"
+# )
+
+# print("Model loaded successfully")
 import torch
 from transformers import AutoProcessor, AutoModelForVision2Seq
 
 MODEL_NAME = "Qwen/Qwen2.5-VL-7B-Instruct"
 
+model = None
+processor = None
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
-print("Loading processor...")
-processor = AutoProcessor.from_pretrained(
-    MODEL_NAME, 
-    trust_remote_code=True,
-    use_fast=True)  # use_fast to avoid warnings in logs
-
-print("Loading model...")
-model = AutoModelForVision2Seq.from_pretrained(
-    MODEL_NAME,
-    trust_remote_code=True,
-    torch_dtype=torch.float16,
-    device_map="auto"
-)
-
-print("Model loaded successfully")
+
+def get_model():
+    global model, processor, device
+
+    if model is None or processor is None:
+        print("Loading processor...")
+        processor = AutoProcessor.from_pretrained(
+            MODEL_NAME,
+            trust_remote_code=True,
+            use_fast=True
+        )
+
+        print("Loading model...")
+        model = AutoModelForVision2Seq.from_pretrained(
+            MODEL_NAME,
+            trust_remote_code=True,
+            torch_dtype=torch.float16,
+            device_map="auto"
+        )
+
+        print("Model loaded successfully")
+
+    return model, processor, device