Long Video Inference Triggers Token Limit Warning & Repetitive Responses

[tangxiao0]

Hi @chenjoya
During the inference process for long videos, I first receive a friendly reminder about exceeding the model's token limit. After that, the program starts generating repetitive responses instead of valid inference results, making it impossible to complete the long video inference task normally. I'm seeking solutions to this issue.
This is a friendly reminder - the current text generation call will exceed the model's predefined maximum length (32768). Depending on the model, you may observe exceptions, performance degradation, or nothing at all.
Specific Phenomenon: After the above warning appears, the model no longer outputs effective inference results. Instead, it repeatedly echoes previous response segments or loops through meaningless duplicate content.
Looking forward to solutions for the repetitive response issue caused by token limit exceeding. Thank you!

[yahooo-m]

I meet the same question, could the authors explain why this generates?

[yahooo-m]

Have u fixed this problem?