When using demo running Intervl2-8b, the response tend to response the same sentence several times until meets max_token_len

[Arthurwyf]

[BryantKai]

我也遇到了这样的问题

[MasterJanus]

same question +1

[czczup]

This issue is likely due to noisy data in the previous model’s training set, causing occasional repetition of sentences. We've trained the new InternVL 2.5 model to reduce this behavior. I recommend trying it, as it performs more smoothly. Let me know if you'd like more details or help setting it up.