GoogleConceptualCaptioning
GoogleConceptualCaptioning copied to clipboard
Multi-GPU instructions
How do I train the code over multi-GPU setup?
set CUDA_VISIBLE_DEVICES, it will automatically use all the gpus.
What about the batch size? Should we explicitly increase it (or) would the code automatically take care by multiplying specified batch size by # GPUs?
You have to explicityly increase it. The batch size is the total batch size.