Unable to generate summary when initializing model with PyTorch

[yungsinatra0]

Hello, I've tried initializing the model using the provided example in the README:

model = BRIO('Yale-LILY/brio-cnndm-uncased', tok.pad_token_id, is_pegasus=False)

However I've been facing some issues when trying to use it for inference:

1. I keep getting issues that parameters used by .generate() method have value of None. I've tried just putting some default values as shown here. Here's how it looks :

inputs = tokenizer([article], max_length=max_length, return_tensors="pt", truncation=True)
summary_ids = model.generate(inputs["input_ids"],
                                     early_stopping=False,
                                     max_length=1024,
                                     num_beams=1,
                                     num_beam_groups=1)

2. This brings me to my second issue, where the .generate() method does not create any ids. Whenever I try to decode the generated summary, I get the error TypeError: 'NoneType' object is not iterable. When checking the type or content of summary_ids, I get None or <class 'NoneType'>.

Why is this happening? When loading the pre-trained models straight from HF, I don't have any issues but this one does not seem to be working.

[thieugiactu]

It's the problem with transformers's version, just use transformer==4.24.0 or anything bellow would work.