2.sh 1.12 KB
Newer Older
lishen's avatar
lishen committed
1

lijian6's avatar
lijian6 committed
2
# rocSHMEM
3
export ROCSHMEM_GDA_NUM_QPS_DEFAULT_CTX=288
4
export ROCSHMEM_MAX_NUM_CONTEXTS=48
lishen's avatar
lishen committed
5
export ROCSHMEM_ALLOWED_IBV_DEVICES=mlx5_2,mlx5_3,mlx5_4,mlx5_5,mlx5_6,mlx5_7,mlx5_8,mlx5_9
lijian6's avatar
lijian6 committed
6
export ROCSHMEM_HEAP_SIZE=10737418240
lijian6's avatar
lijian6 committed
7
8
9
10
11
12
13
14
15
16
17
18
export ROCSHMEM_TOPO_FILE_FORCE=tests/topo.config

# duSHMEM
export LD_LIBRARY_PATH=/opt/dtk/dushmem/lib:$LD_LIBRARY_PATH
export DEEP_EP_DEVICE_TO_HCA_MAPPING=0:mlx5_2:1,1:mlx5_3:1,2:mlx5_4:1,3:mlx5_5:1,4:mlx5_6:1,5:mlx5_7:1,6:mlx5_8:1,7:mlx5_9:1
export NVSHMEM_SYMMETRIC_SIZE=10737418240

# common
export HIP_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
export PYTHONPATH=$(pwd)

# test
lishen's avatar
lishen committed
19
# torchrun --nproc-per-node=1 --nnodes=2 --node-rank=1 --master-addr="10.16.1.37" --master-port=1234 tests/test_internode.py
lishen's avatar
lishen committed
20
# torchrun --nproc-per-node=1 --nnodes=2 --node-rank=1 --master-addr="10.16.1.37" --master-port=1234 tests/test_low_latency.py
lijian6's avatar
lijian6 committed
21
22
torchrun --nproc-per-node=1 --nnodes=2 --node-rank=1 --master-addr="10.16.1.37" --master-port=1234 tests/test_low_latency_new.py --pressure-test
# torchrun --nproc-per-node=1 --nnodes=2 --node-rank=1 --master-addr="10.16.1.37" --master-port=1234 tests/test_internode.py --test-ll-compatibility