liang_chaoming@huawei.com
67 lines
2.1 KiB
YAML
67 lines
2.1 KiB
YAML
tensorflow_config:
|
||
# 基本参数
|
||
data_url: /home/imagenet_TF/
|
||
batch_size: 32
|
||
# 1p/8p, epoches设为90
|
||
epoches: 1
|
||
# 跑精度时max_train_steps设为None
|
||
max_train_steps: 1000
|
||
epochs_between_evals: 1
|
||
iterations_per_loop: 100
|
||
save_checkpoints_steps: 115200
|
||
|
||
# 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: c73:b02
|
||
|
||
# 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
|
||
|
||
pytorch_config:
|
||
# 基本参数
|
||
data_url: /home/imagenet/
|
||
#跑1p时batch_size为512；2p时为1024;4p时为2048;跑8p时batch_size为4096
|
||
batch_size: 512
|
||
epoches: 90
|
||
# train_and_evaluate、evaluate两种模式
|
||
mode: train_and_evaluate
|
||
ckpt_path: /home/train/result/pt_resnet50/training_job_20200916042624/7/checkpoint_npu7model_best.pth.tar
|
||
# docker 镜像名称:版本号
|
||
docker_image: c73:b02
|
||
# 默认参数1p时为0.2，2p/4p/8p时为2.048
|
||
lr: 0.2
|
||
# 指定 device id, 数量需与 rank_size 相同
|
||
device_group_1p: 0
|
||
device_group_2p: 0 1
|
||
device_group_4p: 0 1 2 3
|
||
|
||
mindspore_config:
|
||
# 基本参数
|
||
# 训练时数据集/home/data/imagenet/train, 评测是数据集是/home/data/imagenet/val
|
||
data_url: /home/data/imagenet/train
|
||
#跑1p/2p/4p/8p时batch_size均为256
|
||
batch_size: 256
|
||
epoches: 5
|
||
pre_trained: None
|
||
save_checkpoint_epochs: 5
|
||
loss_scale: 1024
|
||
# mode：train or evaluate
|
||
mode: train
|
||
# 将训练后生成的ckpt的路径配置在此处
|
||
checkpoint_path: /home/wx933135/benchmark_20200924-benchmark_Alpha/benchmark_20200924-benchmark_Alpha/train/result/ms_resnet50/training_job_20200928154504/2/ckpt_2/resnet-5_625.ckpt
|
||
|
||
# eval_device_id,评测是指定的device id
|
||
eval_device_id: 4
|
||
# docker 镜像名称:版本号
|
||
docker_image: c73:b02
|
||
|
||
# 指定 device id, 数量需与 rank_size 相同
|
||
device_group_1p: 0
|
||
device_group_1p: 0 1
|
||
device_group_1p: 0 1 2 3
|
||
|