tensorflow_config:

    train_batch_size: 32
    training_file_pattern: /home/data/raw_data/tfrecord/train2017*
    resnet_checkpoint: /home/data/raw_data/resnet34_pretrain_model/model.ckpt-28152
    validation_file_pattern: /home/data/raw_data/tfrecord/val2017*
    val_json_file: /home/data/raw_data/annotations/instances_val2017.json
    eval_batch_size: 32
    num_epochs: 1
    model_dir: result_npu
    max_steps: 432000

    # train_and_eval, train, eval
    runmode: train_and_eval

    # 指定 device id, 多个 id 使用空格分隔, 数量需与 rank_size 相同
    device_group_1p: 0
    device_group_2p: 0 1
    device_group_4p: 0 1 2 3

    # 仅多机执行需要配置: ip1:卡数量1,ip2:卡数量2
    mpirun_ip: 90.90.176.152:8,90.90.176.154:8

    # docker 镜像名称:版本号
    docker_image: mpirun3:latest