exceed the model's predefined maximum length (4096)

[zhoumengbo]

When I use the Qwen2.5-Math-7B model for inference, I get the following information:

This is a friendly reminder - the current text generation call will exceed the model's predefined maximum length (4096). Depending on the model, you may observe exceptions, performance degradation, or nothing at all.

What is the context length and maximum number of output tokens for this model?

[LaoWangGB]

how to expand the maximum length?

[1998v7]

Sorry to disturb you. Did you reproduce the results of Qwen2.5 math base models provided by the paper?
I only achieved ~70% acc on Gsm8K dataset, which is largely inconsistent with that in the paper.