I have a question regarding the configuration for downstream task training. Specifically, I'm unsure about the number of nodes required to reproduce the results. In the configuration files, such as epic.json, the GPU is set to 8, but in trainer.sh, the srun command is set with N8 and n_gpu:8. Could you please provide clarification on this?
Hello,
I have a question regarding the configuration for downstream task training. Specifically, I'm unsure about the number of nodes required to reproduce the results. In the configuration files, such as
epic.json
, the GPU is set to 8, but intrainer.sh
, thesrun
command is set withN8
andn_gpu:8
. Could you please provide clarification on this?Thank you!