This document presents performance benchmarks for Primus-Turbo.
Work in Progress...
python benchmark/ops/training/deep_ep/test_intranode.pyYou should use slurm or any other tools to run the following:
export NNODES=
export NODE_RANK=
export MASTER_ADDR=
export MASTER_PORT=
torchrun --nproc_per_node 1 --nnodes "${NNODES}" -node_rank "${NODE_RANK}" --master_addr "${MASTER_ADDR}" --master_port "${MASTER_PORT}" benchmark/ops/training/deep_ep/test_internode.py