|
@@ -16,8 +16,7 @@ if [[ $WORLD_SIZE == 1 ]]; then
|
|
|
--root ${DATASET_ROOT} \
|
|
--root ${DATASET_ROOT} \
|
|
|
--model ${MODEL} \
|
|
--model ${MODEL} \
|
|
|
--batch_size ${BATCH_SIZE} \
|
|
--batch_size ${BATCH_SIZE} \
|
|
|
- --resume ${RESUME} \
|
|
|
|
|
- --fp16
|
|
|
|
|
|
|
+ --resume ${RESUME}
|
|
|
elif [[ $WORLD_SIZE -gt 1 && $WORLD_SIZE -le 8 ]]; then
|
|
elif [[ $WORLD_SIZE -gt 1 && $WORLD_SIZE -le 8 ]]; then
|
|
|
python -m torch.distributed.run --nproc_per_node=${WORLD_SIZE} --master_port ${MASTER_PORT} train.py \
|
|
python -m torch.distributed.run --nproc_per_node=${WORLD_SIZE} --master_port ${MASTER_PORT} train.py \
|
|
|
--cuda \
|
|
--cuda \
|
yjh0410