Wandb-Local

Server

1. Launch Server

python3 serve.py --project <project name> --port <port>

• 建立名稱為 <user name>-wandb-<project name> container
• 預設的存儲目錄是 /volume/share/wandb-local/volume/<project name>

若要使用預設路徑外的儲存目錄，可以指定 --volume 參數，例如

python3 serve.py --project <project name> --port <port> --volume /volume/share/wandb-local/volume/trade-richardwang/

2. Registration

1. 瀏覽器打開 http://<host ip>:<port>
   • <host ip> 為當初執行 server 啟動程式的機器的 IP，例如 10.11.60.1
2. 點擊 login 並註冊此部屬上的帳戶，建議設置:
   • Full name: Sinopac
   • E-mail: 123@sinopac.com
   • Username: sinopac
   • Pssword: 123@sinopac.com

Client

Configure connection

可在右上角機器人頭像 -> User Settings -> Danger Zone -> API keys -> Reveal 中找到

寫在 dot env 檔案

設置於 .env 檔案

WANDB_BASE_URL=http:/<host ip>:<port>
WANDB_API_KEY=<api-key>

寫在 Python 檔案

os.environ["WANDB_BASE_URL"] = "http:/<host ip>:<port>"
os.environ["WANDB_API_KEY"] = "<api-key>"

Best practices

Programmatically cleanup cache

wandb use ./cache/local to save artifacts even they are uploaded to server.

c = wandb.wandb_sdk.wandb_artifacts.get_artifacts_cache()
c.cleanup(int(10e9)) # up to 10GB cache

Do not log system metrics

System metrics are frequent and many, which can put a lot of pressure on a local server.

wandb.init(
    ...
    settings=wandb.Settings(
        _disable_stats=True,
        _disable_machine_info=True,
    )
)