forked from jiuyuan/CPM-9G-8B
Update README_ENV.md
This commit is contained in:
parent
8095dc0ad4
commit
c74a6cf86f
|
@ -4,7 +4,7 @@
|
|||
## 镜像加载
|
||||
### rootless 启动
|
||||
允许用户在不影响主机系统的情况下运行应用程序和服务,并且可以轻松地共享和分发环境
|
||||
```js
|
||||
```shell
|
||||
srun -p gpu1 --nodelist=g2001 -N 1 -n 8 -c 8 --gres=gpu:8 --pty bash
|
||||
module load rootless-docker/default
|
||||
```
|
||||
|
@ -13,7 +13,7 @@ module load rootless-docker/default
|
|||
start_rootless_docker.sh运行成功的话,此时执行docker ps可以看到当前没有正在运行的容器,如果有正在运行的容器,说明rootless模式没有启动成功,请联系管理员。
|
||||
|
||||
### 加载镜像
|
||||
```js
|
||||
```shell
|
||||
docker load -i cpmlive-flash-0.0.4.tar
|
||||
docker tag [IMAGE_ID] cpmlive-flash:0.0.4
|
||||
```
|
||||
|
@ -22,12 +22,11 @@ docker tag [IMAGE_ID] cpmlive-flash:0.0.4
|
|||
|
||||
### 启动容器
|
||||
```
|
||||
docker run -it -d -v [HOST_PATH1]:[DOCKER_PATH1] -v [HOST_PATH2]:[DOCKER_PATH2] --gpus all --shm-size=4g --sh cpmlive-flash:0.0.4 bash非rootless 启动
|
||||
docker run -it -d -v [HOST_PATH1]:[DOCKER_PATH1] -v [HOST_PATH2]:[DOCKER_PATH2] --gpus all --shm-size=4g --sh cpmlive-flash:0.0.4 bash
|
||||
```
|
||||
|
||||
如果有docker权限、且rootless执行错误的情况下,可以尝试下非rootless启动
|
||||
加载镜像同rootles过程
|
||||
|
||||
## 非rootless 启动
|
||||
### 启动容器
|
||||
```
|
||||
docker run -it -d -v [HOST_PATH1]:[DOCKER_PATH1] -v [HOST_PATH2]:[DOCKER_PATH2] --gpus all --network host --shm-size=4g cpmlive-flash:0.0.4 bash
|
||||
|
@ -39,19 +38,19 @@ docker run -it -d -v [HOST_PATH1]:[DOCKER_PATH1] -v [HOST_PATH2]:[DOCKER_PATH2]
|
|||
- --network host: 这个选项用于让容器共享主机网络命名空间,使容器可以直接访问主机上的网络接口和服务;
|
||||
- --shm-size 容器的share memory,根据主机的情况设置,如果训练推理需要的内存比较多,可以增大share memory值;
|
||||
### 进入容器
|
||||
```
|
||||
```shell
|
||||
docker exec -it [CONTAINER_ID] bash
|
||||
```
|
||||
### 退出容器
|
||||
```
|
||||
```shell
|
||||
Ctrl+d
|
||||
```
|
||||
### 删除容器
|
||||
```
|
||||
```shell
|
||||
docker stop [CONTAINER_ID]
|
||||
```
|
||||
### 查看正在运行容器
|
||||
```
|
||||
```shell
|
||||
docker ps
|
||||
```
|
||||
|
||||
|
@ -60,10 +59,18 @@ docker ps
|
|||
|
||||
1. 使用python 3.8.10创建conda环境
|
||||
```shell
|
||||
conda create -n cpm-9g python=3.8.10 # 安装Pytorch
|
||||
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia #安装BMTrain
|
||||
pip install bmtrain==0.2.3.post2 #4. 安装flash-attn
|
||||
pip install flash-attn==2.0.8 #5. 安装其他依赖包
|
||||
conda create -n cpm-9g python=3.8.10
|
||||
|
||||
1.安装Pytorch
|
||||
conda install pytorch==1.13.1 torchvision==0.14.1 torchaudio==0.13.1 pytorch-cuda=11.6 -c pytorch -c nvidia
|
||||
|
||||
2. 安装BMTrain
|
||||
pip install bmtrain==0.2.3.post2
|
||||
|
||||
3. 安装flash-attn
|
||||
pip install flash-attn==2.0.8
|
||||
|
||||
4. 安装其他依赖包
|
||||
pip install einops
|
||||
pip install pytrie
|
||||
```
|
||||
|
@ -75,7 +82,9 @@ cuda:11.4-11.6之间都可以
|
|||
### 推理环境配置
|
||||
```js
|
||||
1. 安装nvidia-nccl
|
||||
pip install nvidia-nccl-cu11==2.19.3配置环境变量
|
||||
pip install nvidia-nccl-cu11==2.19.3
|
||||
|
||||
2. 配置环境变量
|
||||
nccl_root=`python -c "import nvidia.nccl;import os; print(os.path.dirname(nvidia.nccl.__file__))"`
|
||||
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$nccl_root/lib
|
||||
echo $LD_LIBRARY_PATH2. 安装LibCPM
|
||||
|
|
Loading…
Reference in New Issue