迁移到下游任务，训练不收敛问题

[CodeStarting-design]

感觉您杰出的工作。 我尝试将您在论文中所提出的聚焦线性注意力模块，迁移到下游的去雾任务中，但是在模型的训练过程中，出现模型参数为nan的问题，然而使用swin替换FLatten模块则不会出现这样的问题，希望能够得到您的解答！

[tian-qing001]

Hi @CodeStarting-design, you can consider turning off amp, which sometimes leads to nan on certain devices. Moreover, setting grad_clip may also helps.

[tian-qing001]

Hi @CodeStarting-design, we have fixed the numerical instability problem. Now the models can be trained with auto mixed precision (amp) in downstream tasks without encountering nan.