search

researchmm / TracKit

[ECCV'20] Ocean: Object-aware Anchor-Free Tracking
MIT License
613 stars 97 forks source link

运行tune_tpe提示CUDA out of memory #51

Closed dasdassssss closed 3 years ago

dasdassssss commented 3 years ago

您好，我用一张GPU尝试tune_tpe，跑了几个小时候显示：RuntimeError: CUDA out of memory. Tried to allocate 90.00 MiB (GPU 0; 10.76 GiB total capacity; 6.31 GiB already allocated; 22.12 MiB free; 6.36 GiB reserved in total by PyTorch) 请问，对于tune tpe一张显卡是否不够用？

JudasDie commented 3 years ago

您好，我用一张GPU尝试tune_tpe，跑了几个小时候显示：RuntimeError: CUDA out of memory. Tried to allocate 90.00 MiB (GPU 0; 10.76 GiB total capacity; 6.31 GiB already allocated; 22.12 MiB free; 6.36 GiB reserved in total by PyTorch) 请问，对于tune tpe一张显卡是否不够用？

一张卡比较久，一般是4-8.你可以把trails调小点。