I'm trying to set up a pre-trained model from huggingface. This is the start of my code:
from transformers import pipeline, TFAutoModelForSequenceClassification
model_name = "microsoft/deberta-base-mnli"
text_classifier = pipeline("text-classification", model=model_name)
model = TFAutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2, ignore_mismatched_sizes=True, from_pt=True)
But already for this last line I get the following error:
Traceback (most recent call last):
File "/home/liv/PycharmProjects/ZSTC_for_MHC/models/baselines/single_task.py", line 29, in <module>
model = TFAutoModelForSequenceClassification.from_pretrained(model_name, num_labels=2, ignore_mismatched_sizes=True,
File "/home/liv/PycharmProjects/ZSTC_for_MHC/.venv/lib/python3.9/site-packages/transformers/models/auto/auto_factory.py", line 446, in from_pretrained
return model_class.from_pretrained(pretrained_model_name_or_path, *model_args, config=config, **kwargs)
File "/home/liv/PycharmProjects/ZSTC_for_MHC/.venv/lib/python3.9/site-packages/transformers/modeling_tf_utils.py", line 1972, in from_pretrained
return load_pytorch_checkpoint_in_tf2_model(model, resolved_archive_file, allow_missing_keys=True)
File "/home/liv/PycharmProjects/ZSTC_for_MHC/.venv/lib/python3.9/site-packages/transformers/modeling_tf_pytorch_utils.py", line 122, in load_pytorch_checkpoint_in_tf2_model
logger.info(f"PyTorch checkpoint contains {sum(t.numel() for t in pt_state_dict.values()):,} parameters")
File "/home/liv/PycharmProjects/ZSTC_for_MHC/.venv/lib/python3.9/site-packages/transformers/modeling_tf_pytorch_utils.py", line 122, in <genexpr>
logger.info(f"PyTorch checkpoint contains {sum(t.numel() for t in pt_state_dict.values()):,} parameters")
AttributeError: 'dict' object has no attribute 'numel'
I'm not calling numel myself and I don't even know which dictionary. Furthermore, I want to use TensorFlow for this, so I don't know if it's alright that it shows a PyTorch error. Anyone know how to fix this? Google didn't provide any solutions.