codes/README_with_internal.md

Build And Deploy

###1.环境要求 需要docker

###2.拉取镜像 算法镜像（13.1G）: sudo docker pull base.image.docker:5000/python/face2face_train:0.0.2_sdk
算法镜像（13.9G）: sudo docker pull base.image.docker:5000/python/face2face_train:0.0.1_sdk_slim
web镜像（550M）: sudo docker pull base.image.docker:5000/python/face2face_train:0.0.2_web

###3.创建共享目录及配置 ####1.创建共享目录 sudo mkdir -p /data/face2face/web /data/face2face/sdk ####2.web配置： /data/face2face/web/config.ini

[path]
log=/code/data/log
temp=/code/data/temp
file_name=face2face_web.log

[face2face]
submit_url=http://<实际宿主机>:8383/easy/submit
query_url=http://<实际宿主机>:8383/easy/query

####3.sdk配置： /data/face2face/sdk/config.ini

[log]
log_dir = /code/data/log
log_file = dh.log

[http_server]
server_ip = 0.0.0.0
server_port = 8383

[temp]
temp_dir = /code/data/temp
clean_switch = 1

[result]
result_dir = /code/data/result
clean_switch = 0

[digital]
batch_size = 4

###4.启动镜像 算法镜像:

sudo  docker run -v /data/face2face:/code/data -v /data/face2face/sdk/config.ini:/code/config/config.ini -e PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512 \
       --gpus all --shm-size 8G -itd --name facesdk  -p 8383:8383  \
       base.image.docker:5000/python/face2face_train:0.0.3_sdk python /code/app_local.py 

算法slim镜像:

sudo  docker run -v /data/face2face:/code/data -v /data/face2face/sdk/config.ini:/code/config/config.ini -e PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512 \
       --gpus all --shm-size 8G -itd --name facesdk  -p 8383:8383  \
       base.image.docker:5000/python/face2face_train:0.0.1_sdk_slim python /code/app_local.py 

web镜像:

docker run -v /data/face2face:/code/data -v /data/face2face/web/config.ini:/code/config.ini \
       -itd --name faceweb  -p 7862:7862 \
       base.image.docker:5000/python/face2face_train:0.0.2_web python -m gradio_web.gradio_test
    

###5.验证 打开浏览器，输入 http://127.0.0.1:7862 即可打开合成页面 ###6.输出目录 /data/face2face/result 页面也可以直接下载 ###7.停止服务 sudo docker stop faceweb && sudo docker rm faceweb sudo docker stop facesdk && sudo docker rm facesdk

遇到的问题及解决方案

Q:docker run 提示 libnvidia-ml.so.1：file exists: unknown
A:需要把cuda相关数据删除:
a.删除/etc/docker/daemon.json中与nvidia相关的内容，其中删除default-runtime和runtimes两个key对应的数据

{
    "default-runtime": "nvidia",  # 需要删除
    "runtimes": {                 # 需要删除
        "nvidia": {               # 需要删除
            "path": "nvidia-container-runtime",   # 需要删除
            "runtimeArgs": []     # 需要删除
        }                         # 需要删除
    },                            # 需要删除
    "registry-mirrors": ["https://nq12ot3x.mirror.aliyuncs.com"],
    "insecure-registries":["base.image.docker:5080","base.image.docker:5000"],
    "data-root": "/home/docker",
    "log-driver":"json-file",
    "log-opts":{ "max-size" :"100m","max-file":"10"}
}
    

b.从当前image中构建新image同时删除libnvidia-ml.so.1、libcuda.so.1和libcudadebugger.so.1三个文件

FROM <当前imageId>
RUN rm -rf /usr/lib/x86_64-linux-gnu/libnvidia-ml.so.1 /usr/lib/x86_64-linux-gnu/libcuda.so.1 /usr/lib/x86_64-linux-gnu/libcudadebugger.so.1