#!/bin/bash
# param 1 is master addr
# param 2 is layer_list
# sh run_dis_llama31_70B.sh 90.90.97.217 4,4 2 2

MASTER_ADDR=$1

echo "master node is $1"
echo "layer list is $2"
echo "TP is $3"
echo "PP is $4"
echo "start master node traning"

source /home/anaconda3/bin/activate llama3

nohup bash examples/mcore/mixtral/pretrain_mixtral_8x7b_ptd_set_layers.sh $1 $2 $3 $4 0 > master_node.log 2>&1 &

sshpass -p HwA3@2025 ssh root@90.90.97.65  << EOF

echo "start nomaster node 246 traning."

cd /home/c00444317/modellink/MindSpeed-LLM-master

source /home/anaconda3/bin/activate llama3

nohup bash examples/mcore/mixtral/pretrain_mixtral_8x7b_ptd_set_layers.sh $1 $2 $3 $4 1 > 246_node.log 2>&1 &

EOF