集合通信与分布式训练

CM_CHIEF_IP
CM_CHIEF_PORT
CM_CHIEF_DEVICE
CM_WORKER_SIZE
CM_WORKER_IP
HCCL_INTRA_PCIE_ENABLE
HCCL_INTRA_ROCE_ENABLE
HCCL_CONNECT_TIMEOUT
HCCL_WHITELIST_FILE
HCCL_WHITELIST_DISABLE
HCCL_IF_IP
HCCL_IF_BASE_PORT
HCCL_EXEC_TIMEOUT
HCCL_RDMA_TC
HCCL_RDMA_SL
HCCL_RDMA_TIMEOUT
HCCL_RDMA_RETRY_CNT
HCCL_SOCKET_IFNAME
HCCL_P2P_DISABLE

父主题： 常用环境变量说明