search

motional / nuplan-devkit

The devkit of the nuPlan dataset.
https://www.nuplan.org
Other
673 stars 129 forks source link

what's the difference between the first and second case in initialize_ray in nuplan/planning/utils/multithreading/worker_ray.py #263

Closed yaqlee closed 1 year ago

yaqlee commented 1 year ago

I tried to train model on multi instances wit Ray Distributed. I felt confused about the usage of nuplan/planning/script/config/common/worker/ray_distributed.yaml, should I set the master_node-ip？ To the point, what's the difference between the first and second case in initialize_ray in nuplan/planning/utils/multithreading/worker_ray.py？as shown in master node ip case

yaqlee commented 1 year ago

in this line, the num of node was set to 1, but if I want to train on multi machines, the num of nodes should be as same as the number of machines. how does it work？

patk-motional commented 1 year ago

Hi @yaqlee,

We do not use ray within nuPlan framework to manage multi-node training. The code is there because we have attempted to use ray in the past. However, internally we have pivoted to a custom solution that we cannot share with you publicly. Therefore, we cannot help you with this issue unfortunately.