【踩坑日记6】如何使用vscode远程调试accelerate launch语句

问题描述

想debug语句：

accelerate launch --multi_gpu --mixed_precision 'no' --machine_rank 0  --main_process_ip 222.201.144.234  --main_process_port 7234  --num_machines 1 --num_processes 1 ./diff_prompter/ppo_prompter.py --data ./data_path --gpt_path ./supervised_finetuned_gpt_path --trl_config ./diff_prompter/configs/ppo_config_a100_coco_bsz256_kl0.2.yml --checkpoint_dir ./ckpt_dir 

问题解决

将launch.json配置文件添加：

 "module": "accelerate.commands.launch"

即改为：

{
    // 使用 IntelliSense 了解相关属性。 
    // 悬停以查看现有属性的描述。
    // 欲了解更多信息，请访问: https://go.microsoft.com/fwlink/?linkid=830387
    "version": "0.2.0",
    "configurations": [
        {
            "name": "promptist",
            "type": "debugpy",
            "request": "launch",
            "module": "accelerate.commands.launch",          //调试accelerate launch
            "args": [
                "--multi_gpu", 
                "--mixed_precision",      "no",
                "--machine_rank",         "0", 
                "--main_process_ip",      "222.201.144.234", 
                "--main_process_port",    "7234", 
                "--num_machines",         "1", 
                "--num_processes",        "1", 
                "./diff_prompter/ppo_prompter.py", 
                "--data",                 "./data_path", 
                "--gpt_path",             "./supervised_finetuned_gpt_path", 
                "--trl_config",           "./diff_prompter/configs/ppo_config_a100_coco_bsz256_kl0.2.yml", 
                "--checkpoint_dir",       "./ckpt_dir",
            ],                                                           //需要输入的参数
            "cwd": "/mnt/lab/***/code/promptist",    //当前调试的目录
            "console": "integratedTerminal"
        }
    ]
}

参考

vscode连接服务器进行远程开发调试

让vscode在远程连接服务器时候如本地般顺滑地debug

How can I debug on vscode #400

vscode调试pytorch分布式训练