search

232525 / PureT

Implementation of 'End-to-End Transformer Based Model for Image Captioning' [AAAI 2022]
63 stars 12 forks source link

some questions about PureT_SCST #4

Closed zml110120 closed 2 years ago

zml110120 commented 2 years ago

Could you tell me which kind of GPU and how many you has uesd? I use V100-16GB and there is OOM error

232525 commented 2 years ago

V100-32GB, batch size of 10, single GPU is enough.

232525 commented 2 years ago

Actually 16G is enough, I try to train under SCST with batch size of 10, it just cost 5000M - 6000M CUDA memory.

232525 commented 2 years ago

Actually 16G is enough, I try to train under SCST with batch size of 10, it just cost 5000M - 6000M CUDA memory.

Oh, I am sorry, it is wrong, I trained the model without the backbone model of SwinT, the full model may need more CUDA memory.