Change to better import dependency

src/models/ocr_model/model/TexTeller.py

@@ -1,5 +1,4 @@
 from pathlib import Path
-from optimum.onnxruntime import ORTModelForVision2Seq
 
 from ...globals import (
     VOCAB_SIZE,
@@ -32,6 +31,7 @@ class TexTeller(VisionEncoderDecoderModel):
             if not use_onnx:
                 return VisionEncoderDecoderModel.from_pretrained(cls.REPO_NAME)
             else:
+                from optimum.onnxruntime import ORTModelForVision2Seq
                 use_gpu = True if onnx_provider == 'cuda' else False
                 return ORTModelForVision2Seq.from_pretrained(cls.REPO_NAME, provider="CUDAExecutionProvider" if use_gpu else "CPUExecutionProvider")
         model_path = Path(model_path).resolve()