未验证 提交 ce0e6fe5 编写于 作者: C Chengmo 提交者: GitHub

fix gpu default card & doc (#79)

* fix gpu default card & doc
上级 37dfcd3b
......@@ -18,7 +18,7 @@
| class | string | train(默认) / infer / local_cluster / cluster | 是 | 指定运行runner的类别(单机/分布式, 训练/预测) |
| device | string | cpu(默认) / gpu | 否 | 程序执行设备 |
| fleet_mode | string | ps(默认) / pslib / collective | 否 | 分布式运行模式 |
| selsected_gpus | string | "0"(默认) | 否 | 程序运行GPU卡号,若以"0,1"的方式指定多卡,则会默认启用collective模式 |
| selected_gpus | string | "0"(默认) | 否 | 程序运行GPU卡号,若以"0,1"的方式指定多卡,则会默认启用collective模式 |
| worker_num | int | 1(默认) | 否 | 参数服务器模式下worker的数量 |
| server_num | int | 1(默认) | 否 | 参数服务器模式下server的数量 |
| distribute_strategy | string | async(默认)/sync/half_async/geo | 否 | 参数服务器模式下训练模式的选择 |
......
......@@ -113,7 +113,7 @@ def get_engine(args, running_config, mode):
engine_class = ".".join(["runner", mode, "class"])
engine_device = ".".join(["runner", mode, "device"])
device_gpu_choices = ".".join(["runner", mode, "device", "selected_gpus"])
device_gpu_choices = ".".join(["runner", mode, "selected_gpus"])
engine = running_config.get(engine_class, None)
if engine is None:
......@@ -135,7 +135,7 @@ def get_engine(args, running_config, mode):
print(
"not find selected_gpus be specified in yaml, set `0` as default"
)
selected_gpus = ["0"]
selected_gpus = "0"
else:
print("selected_gpus {} will be specified for running".format(
selected_gpus))
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册