Should this be after_texts[j]?

[ni-eminen]

LMDrive/LAVIS/lavis/models/drive_models/drive.py

Line 344 in 43fc2e9

p_after_tokens = self.llm_tokenizer(before_texts[j], return_tensors="pt", add_special_tokens=False).to(image_embeds.device)

[deepcs233]

Yes, you are right! But the current code didn't turn on the 'use_extra_prompt' flag, so the bug didn't affect the performance.

[ni-eminen]

Is this issue cloned from another repository, or unique to this project? I submitted a pr fixing this in this repo.