🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.
thanks for sharing. i use otter_video.py to test luodian/OTTER-MPT1B-RPJama-Init, but find that output is empty,and the decdoer id is large than model.text_tokenizer.vocab_size, can you give me some explanation？
thanks for sharing. i use otter_video.py to test luodian/OTTER-MPT1B-RPJama-Init, but find that output is empty,and the decdoer id is large than model.text_tokenizer.vocab_size, can you give me some explanation？