try using NougatProcessor using custom transformer version

handler.py

@@ -1,13 +1,13 @@
 from io import BytesIO
 from typing import Dict, Any
-from transformers import AutoProcessor, VisionEncoderDecoderModel
+from transformers import NougatProcessor, VisionEncoderDecoderModel
 from transformers.image_utils import base64
 from PIL import Image
 import torch
 
 class EndpointHandler():
     def __init__(self, path="facebook/nougat-base") -> None:
-        self.processor = AutoProcessor.from_pretrained(path)
+        self.processor = NougatProcessor.from_pretrained(path)
         self.model = VisionEncoderDecoderModel.from_pretrained(path)
 
         self.device = "cuda" if torch.cuda.is_available() else "cpu"

preprocessor_config.json

@@ -11,13 +11,13 @@
     0.456,
     0.406
   ],
-  "image_processor_type": "AutoProcessor",
+  "image_processor_type": "NougatImageProcessor",
   "image_std": [
     0.229,
     0.224,
     0.225
   ],
-  "processor_class": "AutoProcessor",
+  "processor_class": "NougatProcessor",
   "resample": 2,
   "rescale_factor": 0.00392156862745098,
   "size": {

tokenizer_config.json

@@ -195,9 +195,9 @@
   "pad_token": "<pad>",
   "pad_token_type_id": 0,
   "padding_side": "right",
-  "processor_class": "AutoProcessor",
+  "processor_class": "NougatProcessor",
   "stride": 0,
-  "tokenizer_class": "AutoTokenizer",
+  "tokenizer_class": "NougatTokenizer",
   "truncation_side": "right",
   "truncation_strategy": "longest_first",
   "unk_token": "<unk>",