zai-org

/

GLM-OCR

like 1.03k

Follow

Z.ai 9.38k

Is there anyone who got error when loading AutoProcessor?
[versions]
"torch==2.9.1" (cpu only)
"transformers==5.1.0"

---> 20 processor = AutoProcessor.from_pretrained("zai-org/GLM-OCR")
     21 model = AutoModelForImageTextToText.from_pretrained(
     22     pretrained_model_name_or_path=MODEL_PATH,
     23     torch_dtype="auto",
     24     device_map="auto",
     25 )
     26 inputs = processor.apply_chat_template(
     27     messages,
     28     tokenize=True,
   (...)     31     return_tensors="pt"
     32 ).to(model.device)

File c:\Users\flash\projects\cpuinfer-convert\.venv\Lib\site-packages\transformers\models\auto\processing_auto.py:398, in AutoProcessor.from_pretrained(cls, pretrained_model_name_or_path, **kwargs)
    394     return processor_class.from_pretrained(
    395         pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
    396     )
    397 elif processor_class is not None:
--> 398     return processor_class.from_pretrained(
    399         pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
    400     )
    401 # Last try: we use the PROCESSOR_MAPPING.
    402 elif type(config) in PROCESSOR_MAPPING:

File c:\Users\flash\projects\cpuinfer-convert\.venv\Lib\site-packages\transformers\processing_utils.py:1402, in ProcessorMixin.from_pretrained(cls, pretrained_model_name_or_path, cache_dir, force_download, local_files_only, token, revision, **kwargs)
   1400 # Get processor_dict first so we can use it to instantiate non-tokenizer sub-processors
   1401 processor_dict, instantiation_kwargs = cls.get_processor_dict(pretrained_model_name_or_path, **kwargs)
-> 1402 args = cls._get_arguments_from_pretrained(pretrained_model_name_or_path, processor_dict, **kwargs)
   1403 return cls.from_args_and_dict(args, processor_dict, **instantiation_kwargs)

File c:\Users\flash\projects\cpuinfer-convert\.venv\Lib\site-packages\transformers\processing_utils.py:1523, in ProcessorMixin._get_arguments_from_pretrained(cls, pretrained_model_name_or_path, processor_dict, **kwargs)
   1520 elif is_primary:
   1521     # Primary non-tokenizer sub-processor: load via Auto class
   1522     auto_processor_class = MODALITY_TO_AUTOPROCESSOR_MAPPING[sub_processor_type]
-> 1523     sub_processor = auto_processor_class.from_pretrained(
   1524         pretrained_model_name_or_path, subfolder=subfolder, **kwargs
   1525     )
   1526     args.append(sub_processor)
   1528 elif sub_processor_type in processor_dict:
   1529     # Additional non-tokenizer sub-processor: instantiate from config in processor_dict

File c:\Users\flash\projects\cpuinfer-convert\.venv\Lib\site-packages\transformers\models\auto\video_processing_auto.py:342, in AutoVideoProcessor.from_pretrained(cls, pretrained_model_name_or_path, *inputs, **kwargs)
    338     video_processor_class_inferred = image_processor_class.replace("ImageProcessor", "VideoProcessor")
    340     # Some models have different image processors, e.g. InternVL uses GotOCRImageProcessor
    341     # We cannot use GotOCRVideoProcessor when falling back for BC and should try to infer from config later on
--> 342     if video_processor_class_from_name(video_processor_class_inferred) is not None:
    343         video_processor_class = video_processor_class_inferred
    344 if "AutoImageProcessor" in config_dict.get("auto_map", {}):

File c:\Users\flash\projects\cpuinfer-convert\.venv\Lib\site-packages\transformers\models\auto\video_processing_auto.py:96, in video_processor_class_from_name(class_name)
     94 def video_processor_class_from_name(class_name: str):
     95     for module_name, extractors in VIDEO_PROCESSOR_MAPPING_NAMES.items():
---> 96         if class_name in extractors:
     97             module_name = model_type_to_module_name(module_name)
     99             module = importlib.import_module(f".{module_name}", "transformers.models")

TypeError: argument of type 'NoneType' is not iterable