BertLMHeadModel has no attribute 'generate'

[allisonandreyev]

Tag2Text inference crashes with AttributeError: 'BertLMHeadModel' object has no attribute 'generate'. Repro: results = inference(image_tensor, model, 'None') which reaches tag2text.py:355 and calls self.text_decoder.generate(...). In my env (macOS, Python 3.13, Transformers 4.x), BertLMHeadModel does not expose generate. Is there a required Transformers version, or should the decoder be a generation-capable class like AutoModelForCausalLM or a seq2seq model? If a version pin is intended, please specify. I can test a small patch that guards this call with hasattr(self.text_decoder, "generate").