script/start_node2_tp32_pytorch.sh

#!/bin/bash
# Node 2 - 192.168.2.123

# 日志目录
LOG_DIR=/data/models/zyl/glm/logs
LOG_FILE=${LOG_DIR}/node2_$(date +%Y%m%d_%H%M%S).log
mkdir -p ${LOG_DIR}

# MCCL 多节点通信配置
export MCCL_HC_PLUGIN=/opt/maca-3.3.0/lib/libmxccl_21605plugin.so
export MCCL_EXT_CCL_ENABLE=1
export NCCL_FORCE_USE_MN_MIX=1
export MCCL_SOCKET_IFNAME=inbond1
export NCCL_SOCKET_IFNAME=inbond1
export GLOO_SOCKET_IFNAME=inbond1
export MCCL_DEBUG=WARN
export MCCL_PROTO=LL
export MCCL_P2P_LL_THRESHOLD=1073741824
export MCCL_IB_DISABLE=1
export NCCL_IB_DISABLE=1
export NCCL_DEBUG=WARN
export MASTER_ADDR=192.168.2.121
export MASTER_PORT=29501

echo "[$(date)] Starting node2..." | tee ${LOG_FILE}
echo "[$(date)] Log file: ${LOG_FILE}" | tee -a ${LOG_FILE}
echo "[$(date)] Use 'tail -f ${LOG_FILE}' to monitor" | tee -a ${LOG_FILE}

nohup env FLAGGEMS_VENDOR=metax \
  USE_FLAGGEMS=0 \
  VLLM_FL_OOT_ENABLED=0 \
  VLLM_FL_PREFER_ENABLED=0 \
  vllm serve --headless /data/models/GLM-5.2-0610-Provider-BF16 \
  --host 0.0.0.0 \
  --port 8000 \
  --served-model-name glm52 \
  --tensor-parallel-size 32 \
  --nnodes 4 \
  --node-rank 2 \
  --master-addr 192.168.2.121 \
  --enforce-eager \
  --max-model-len 32768 \
  >> ${LOG_FILE} 2>&1 &

echo "[$(date)] Node2 PID: $!" | tee -a ${LOG_FILE}
echo ""
echo "=== Node2 started in background ==="
echo "Monitor: tail -f ${LOG_FILE}"