Update app.py

app.py

@@ -6,12 +6,16 @@ from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoPro
 from qwen_vl_utils import process_vision_info
 import cv2
 
-# Define the directory containing your downloaded model files
-model_directory = "/path/to/your/local/model/files"
 
-# Load model weights directly from local files
-model = Qwen2VLForConditionalGeneration(config=model_directory)
-model.load_state_dict(torch.load(f"{model_directory}/model.safetensors"))
+# Đường dẫn đến thư mục chứa các file .safetensors và các file cấu hình
+model_directory = "THP2903/erax_llm"
+
+# Load model từ các file .safetensors shards trong thư mục đã tải
+model = Qwen2VLForConditionalGeneration.from_pretrained(
+    model_directory,
+    torch_dtype=torch.bfloat16,  # Sử dụng torch.float16 nếu cần tiết kiệm bộ nhớ
+    device_map="auto"
+)
 
 # Load tokenizer from local files
 tokenizer = AutoTokenizer(
@@ -21,7 +25,7 @@ tokenizer = AutoTokenizer(
 )
 
 # Load processor directly from config
-processor_config = torch.load(f"{model_directory}/processor_config.json")
+processor_config = torch.load(f"{model_directory}/preprocessor_config.json")
 processor = AutoProcessor.from_config(processor_config)
 
 # Set generation configuration directly