udpate

finetune/Dockerfile

@@ -376,8 +376,6 @@ cd ${STAGE_DIR}/DeepSpeed
 git checkout . 
 git checkout v${DEEPSPEED_VERSION}
 # python setup.py bdist_wheel
-# DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v dist/deepspeed*.whl --force-reinstall
-# DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v -r requirements/requirements.txt
 # 修改 install.sh 脚本中的 python 解释器路径
 # sed "s|\bpython\b|/opt/conda/envs/${CONDA_ENV_NAME}/bin/python|g" install.sh > install_modified.sh
 # chmod +x ./install_modified.sh
@@ -389,6 +387,9 @@ else
     INSTALL_CMD="./install.sh ${DEEPSPEED_INSTALL_FLAGS}"
 fi
 eval $INSTALL_CMD
+DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v dist/deepspeed*.whl --force-reinstall
+DS_BUILD_OPS=${DS_BUILD_OPS} pip install -v -r requirements/requirements.txt
+pip install numpy==1.22.4 # ImportError: cannot import name 'BUFSIZE' from 'numpy' (/opt/conda/envs/deepspeed/lib/python3.10/site-packages/numpy/__init__.py) wait for fix in numpy=2.0.0
 EOT
 
 # install transformers and flash-attn

finetune/README.md

@@ -59,3 +59,101 @@ Device 7: "NVIDIA H100 80GB HBM3"
   CUDA Capability Major/Minor version number:    9.0
 
 ```
+
+
+## DeepSpeed hostfile 分发
+
+要手动分发 hostfile 并进行分布式安装，你需要以下几个步骤：
+
+1. 准备 hostfile
+确保 hostfile 文件包含所有参与的主机及其配置。
+
+示例 hostfile 内容：
+
+```plaintext
+host1 slots=4
+host2 slots=4
+host3 slots=8
+```
+
+2. 确保 SSH 配置正确
+确保你能够通过 SSH 无密码登录到所有主机。可以使用 ssh-keygen 和 ssh-copy-id 配置 SSH 密钥。
+
+生成 SSH 密钥（如果尚未生成）：
+
+```shell
+ssh-keygen -t rsa
+```
+
+将 SSH 公钥复制到每个主机：
+
+```shell
+ssh-copy-id user@host1
+ssh-copy-id user@host2
+ssh-copy-id user@host3
+```
+
+3. 创建临时目录并复制 wheel 文件
+在所有主机上创建一个临时目录，用于存放分发的 wheel 文件。
+
+```shell
+export PDSH_RCMD_TYPE=ssh
+hosts=$(cat /path/to/your/hostfile | awk '{print $1}' | paste -sd ",")
+tmp_wheel_path="/tmp/deepspeed_wheels"
+
+pdsh -w $hosts "mkdir -pv ${tmp_wheel_path}"
+pdcp -w $hosts dist/deepspeed*.whl ${tmp_wheel_path}/
+pdcp -w $hosts requirements/requirements.txt ${tmp_wheel_path}/
+```
+
+4. 在每个主机上安装 DeepSpeed 和依赖项
+在所有主机上安装 DeepSpeed 和所需的依赖项。
+
+```shell
+pdsh -w $hosts "pip install ${tmp_wheel_path}/deepspeed*.whl"
+pdsh -w $hosts "pip install -r ${tmp_wheel_path}/requirements.txt"
+```
+
+5. 清理临时文件
+安装完成后，删除所有主机上的临时文件。
+
+```shell
+pdsh -w $hosts "rm -rf ${tmp_wheel_path}"
+```
+
+详细步骤
+确保 SSH 配置正确：
+
+```shell
+ssh-keygen -t rsa
+ssh-copy-id user@host1
+ssh-copy-id user@host2
+ssh-copy-id user@host3
+```
+
+创建临时目录并复制文件：
+
+```shell
+export PDSH_RCMD_TYPE=ssh
+hosts=$(cat /path/to/your/hostfile | awk '{print $1}' | paste -sd ",")
+tmp_wheel_path="/tmp/deepspeed_wheels"
+
+pdsh -w $hosts "mkdir -pv ${tmp_wheel_path}"
+pdcp -w $hosts dist/deepspeed*.whl ${tmp_wheel_path}/
+pdcp -w $hosts requirements/requirements.txt ${tmp_wheel_path}/
+```
+
+在所有主机上安装 DeepSpeed 和依赖项：
+
+```shell
+pdsh -w $hosts "pip install ${tmp_wheel_path}/deepspeed*.whl"
+pdsh -w $hosts "pip install -r ${tmp_wheel_path}/requirements.txt"
+```
+
+清理临时文件：
+
+```shell
+pdsh -w $hosts "rm -rf ${tmp_wheel_path}"
+```
+
+通过这些步骤，你可以手动分发 hostfile 并在多个主机上安装 DeepSpeed 和其依赖项。这种方法确保了每个主机的环境配置一致，从而支持分布式训练或部署。

finetune/docker-compose_pytorch2.3.yml

@@ -35,6 +35,7 @@ services:
         CUDA_ARCH_LIST: "80;86;89;90" # for RTX 4090, all : "80;86;89;90"
         SETUPTOOLS_VERSION: "69.5.1"
         DCUTLASS_NVCC_ARCHS: "80;86;89;90;90a" # 90a for H100 GPU 89:GeForce RTX 4090 
+        DEEPSPEED_VERSION: "0.14.3"
         DEEPSPEED_INSTALL_FLAGS: "--allow_sudo --pip_sudo --verbose"
     volumes:
       - ./src:/bbtft