6.1 Hello FV

    What is Hello FV ?

    Hello-FV 主要由python撰寫，讓開發者學習Ailabs's FV framework，Hello-FV使用MNIST作示範。

    Getting started

    請準備一個Ubuntu 20.04的環境，且docker環境已經被安裝好。

    1.請從Pytorch官方下載MNIST資料集，或可經由我們的連結mnist_dataset.zip下載，下載完成請解壓縮，並將解壓縮後的MNIST資料夾放到/var/data下。

    2.請下載Ailabs所提供MNIST的model weight，下載完請確認檔名為`model_weight.ckpt`並放在 `/var/model`底下，完整路徑即`/var/model/model_weight.ckpt`可讀取到MNIST's model weight。

    3.在command line 輸入以下指令啟動Hello-FL已經預先build出來的image。

    docker run -it --name hello-fv --runtime=nvidia \
    -e NVIDIA_VISIBLE_DEVICES=0 \
    -v /var/run/docker.sock:/var/run/docker.sock \
    -v /bin/docker:/bin/docker \
    -v /var/model/model_weight.ckpt:/model_weight.ckpt \
    -v /var/data:/data \
    -v /var:/var \
    -v /var/output:/var/output \
    -v /var/logs:/var/logs \
    registry.corp.ailabs.tw/federated-learning/hello-fv/edge:1.1.1
    我們所提供上述的啟動指令，其中幾個比較重要的參數意義如下
    NVIDIA_VISIBLE_DEVICES=0 : A FV plan will need GPU to do validation, this value is the index of one of your GPU cards. Will
    be 0 if the developer just has one GPU in his edge.

    -v /var/data:/data : This is the path where the container load the datasets from.

    -v /var/model/model_weight.ckpt:/model_weight.ckpt** : The location of model's weight. Put the MNIST's model weight
    downloaded at **/var/model_weight.ckpt **.

    -v /var/output:/var/output** : This is the path where the container output the result of validation (**result.json**).

    -v /var/logs:/var/logs** : This is the path where the container outputs the logs of the developer's container. And
    the progress of validation(**progress.json**) is also located here.

    Copyright © 2022Powered by Bludit