multi gpus training cause out of memory error

[Dufresue]

when i try to run the train_caption.py script like this:

export Data_ROOT=path/to/coco_dataset
python train_caption.py exp.name=caption_rds moel.detector.checkpoint=4ds_detector_path

i encountered some errors like this:

belows are changes i made in coco_config.yml:
ngpus_per_node:2
world_size:2
batch_size:4
num_workers:2

however,when i set the ngpus_per_node:1 world_size:1，it can run properly.

anyone can help? thanks a lot!

[Dufresue]

i believe that is caused by the out of memory. i open the system monitor and found that after running the traincaption.py script, the memory usage continues to increase until it is full, causing an oom error. my computer has 16g memory, is the problem caused by insufficient memory? Or is it because there is a problem in the code that resources are not released in time?