vscode | python | remote-SSH | Debug 配置 + CLIP4Clip实验记录

news2025/1/24 14:34:21

安装Extension

本地安装Remote-SSH、python
远程服务器上安装Python

难点：主机和远程服务器上安装Python扩展失败，可能是网络、代理等原因导致
解决方法：
- 主机在官方网站下载Python扩展：https://marketplace.visualstudio.com/items?itemName=ms-python.python
  主机直接放在vscode的bin目录下并且执行指令code --install-extension ms-python.python-2022.9.11681004.vsix即可
  （细节见https://www.hangge.com/blog/cache/detail_3191.html）
- 服务器的python扩展先使用scp从本地传上去，然后先要对其赋予执行权限，我一开始没有解决就是因为没有赋予权限，我直接chmod 777之后install from vsix即可（chmod +x应该也行）
  
  之后就看到环境了：
  
  现在可以选择自己在服务器的conda进行调试：
  
  价值一天半时间的”权限访问“难题被破解！此时不禁想要听一百遍越权访问加深印象…

之后就要run->add configuration->
launch.json如下：

{
    "version": "0.2",
    "configurations": [
        {
            "name": "Python: Launch",
            "type": "python",
            "request": "launch",
            "program": "${workspaceFolder}/CLIP4Clip/main_task_retrieval.py",
            "args": [
                "--do_train",
                "--num_thread_reader=0",
                "--epochs=5",
                "--batch_size=128",
                "--n_display=50",
                "--train_csv",
                "${env:DATA_PATH}/MSRVTT_train.9k.csv",
                "--val_csv",
                "${env:DATA_PATH}/MSRVTT_JSFUSION_test.csv",
                "--data_path",
                "${env:DATA_PATH}/MSRVTT_data.json",
                "--features_path",
                "${env:DATA_PATH}/MSRVTT_Videos",
                "--output_dir",
                "ckpts/ckpt_msrvtt_retrieval_looseType",
                "--lr",
                "1e-4",
                "--max_words",
                "32",
                "--max_frames",
                "12",
                "--batch_size_val",
                "16",
                "--datatype",
                "msrvtt",
                "--expand_msrvtt_sentences",
                "--feature_framerate",
                "1",
                "--coef_lr",
                "1e-3",
                "--freeze_layer_num",
                "0",
                "--slice_framepos",
                "2",
                "--loose_type",
                "--linear_patch",
                "2d",
                "--sim_header",
                "meanP",
                "--pretrained_clip_name",
                "ViT-B/32"
            ],
            "env": {
                "DATA_PATH": "/mnt/cloud_disk/wf/msrvtt_data"
            },
            "console": "integratedTerminal"
        }
    ]
}

之后出现一个问题就是目前引用env变量在命令行中显示为空，目前不能用这个方式引用所以还得用笨方法，就是挨个复制粘贴。
并且python -m要变成module词段，module与program冲突，需要调整：

{
    "version": "0.2",
    "configurations": [
        {
            "name": "Python: Launch",
            "type": "python",
            "request": "launch",
            "module": "torch.distributed.launch",
            "args": [
                "${workspaceFolder}/CLIP4Clip/main_task_retrieval.py",
                "--do_train",
                "--num_thread_reader=0",
                "--epochs=5",
                "--batch_size=128",
                "--n_display=50",
                "--train_csv",
                "/mnt/cloud_disk/wf/msrvtt_data/MSRVTT_train.9k.csv",
                "--val_csv",
                "/mnt/cloud_disk/wf/msrvtt_data/MSRVTT_JSFUSION_test.csv",
                "--data_path",
                "/mnt/cloud_disk/wf/msrvtt_data/MSRVTT_data.json",
                "--features_path",
                "/mnt/cloud_disk/wf/msrvtt_data/MSRVTT_Videos",
                "--output_dir",
                "ckpts/ckpt_msrvtt_retrieval_looseType",
                "--lr",
                "1e-4",
                "--max_words",
                "32",
                "--max_frames",
                "12",
                "--batch_size_val",
                "16",
                "--datatype",
                "msrvtt",
                "--expand_msrvtt_sentences",
                "--feature_framerate",
                "1",
                "--coef_lr",
                "1e-3",
                "--freeze_layer_num",
                "0",
                "--slice_framepos",
                "2",
                "--loose_type",
                "--linear_patch",
                "2d",
                "--sim_header",
                "meanP",
                "--pretrained_clip_name",
                "ViT-B/32"
            ],
            "console": "integratedTerminal"
        }
    ]
}

之后设置断点调试之后发现这个问题：
在这里插入图片描述
挨个语句调试之后发现出现在某个加载模型的地方，模型的位置防止错误了，远程调试真的好用，可以清晰看到过程的调用栈call stack