值得顺便掌握一下的那些linux指令
命令行能够即时地做很多事情，甚至代表一种简洁而切题的计算机哲学。
为了直接进入正题，省略诸如 cd 这样的广为人知的基础指令，直接看那些显著提升体验的东西。
edit time: 2026-03-10 11:31:56
df
用于查看磁盘空间。使用场景比如训模型会保存很多中间文件，可以这样看看需要清理哪些。
使用效果：
比如 fuse-overlayfs 是这个实例本身的存储， shared-nvme 是它的外接可持久化存储，使用率分别为 50% 和 40% 。
(cs336) root@p-4e088649b993-ackcs-00gjgdno:~# df
Filesystem 1K-blocks Used Available Use% Mounted on
fuse-overlayfs 31457280 15468244 15989036 50% /
/dev/sda3 931347132 85832692 845514440 10% /base
tmpfs 65536 0 65536 0% /dev
tmpfs 528231156 12 528231144 1% /proc/driver/nvidia
tmpfs 528231156 0 528231156 0% /proc/asound
tmpfs 528231156 0 528231156 0% /proc/acpi
tmpfs 65536 0 65536 0% /dev/tty
tmpfs 528231156 0 528231156 0% /proc/scsi
JuiceFS:public-juice3 21474836480 16810363272 4664473208 79% /shared-public
tmpfs 528231156 0 528231156 0% /sys/firmware
tmpfs 528231156 0 528231156 0% /sys/devices/virtual/powercap
/dev/xvdb 308521792 15468280 277308488 6% /etc/hosts
tmpfs 62914560 112 62914448 1% /dev/shm
JuiceFS:shared-juice3 104857600 41665660 63191940 40% /root/shared-nvme
overlay 931347132 85832692 845514440 10% /run/init
tmpfs 105646232 57136 105589096 1% /run/nvidia-persistenced/socket
至于要看哪几个文件最大，可以和 ncdu 这样的工具搭配使用。
grep
查找关键词，关键词可以写正则。
通常和其他指令通过管道搭配使用，简单的用法、强大的功能。
使用效果：
(cs336) root@p-4e088649b993-ackcs-00gjgdno:~# df | grep /dev
/dev/sda3 931347132 85849780 845497352 10% /base
tmpfs 65536 0 65536 0% /dev
tmpfs 65536 0 65536 0% /dev/tty
tmpfs 528231156 0 528231156 0% /sys/devices/virtual/powercap
/dev/xvdb 308521792 15468280 277308488 6% /etc/hosts
tmpfs 62914560 112 62914448 1% /dev/shm
也可以加 -i 参数用于忽略大小写。
ps
ps aux 用于查看当前用户的进程。
但是输出往往很冗长。我只关心我运行的实验，怎么做？
答案很方便：管道加grep！
(cs336) root@p-4e088649b993-ackcs-00gjgdno:~# ps aux | grep uv
root 7482 0.0 0.0 232328 36616 pts/0 Sl+ 11:02 0:00 uv run scripts/batch_sweep.sh --use-wandb --train-data data/tinystories_train.bin --val-data data/tinystories_val.b
root 7512 0.0 0.0 232332 36100 pts/0 Sl+ 11:02 0:00 uv run python -m cs336_basics.train --train-data data/tinystories_train.bin --data-dtype uint16 --vocab-size 10000
root 8012 0.0 0.0 3960 1976 pts/1 S+ 11:27 0:00 grep --color=auto uv
作为对比，以下是原始输出。信息量比较大，不便于我们立即锁定自己需要的东西。
(cs336) root@p-4e088649b993-ackcs-00gjgdno:~# ps aux
USER PID %CPU %MEM VSZ RSS TTY STAT START TIME COMMAND
root 189 0.0 0.0 1128 4 ? S 08:24 0:00 catatonit -P
root 468 0.0 0.0 11868 10472 ? Ss 08:24 0:00 /usr/bin/fuse-overlayfs -o lowerdir=/pcs/image_storage/overlay/l/FYLFLVKOA36HFQA3YPVYA4TOPN:/pcs/image_storage/over
root 730 0.0 0.0 11642352 42724 ? Sl 08:24 0:10 podman start -a -l
root 804 0.0 0.0 140084 2080 ? Ssl 08:24 0:00 /usr/libexec/podman/conmon --api-version 1 -c 21cc623ce89ebf5b23d24fab89813f67368a3a7d85627ea813d278eceae6b5d2 -u 2
root 809 0.0 0.0 4884 3516 ? Ss 08:24 0:00 /bin/bash /run/init
root 4497 0.0 0.0 37064 31664 ? S 08:24 0:01 /base/mambaforge/bin/python /base/mambaforge/bin/supervisord -n
root 4804 0.0 0.0 4752 3404 ? S 08:24 0:00 bash -lc while true; do /base/mambaforge/bin/jupyter lab --ip 0.0.0.0 --port 8888 --no-browser --ServerApp.root_d
root 4805 0.0 0.0 4752 3400 ? S 08:24 0:00 bash -lc while true; do /base/mambaforge/bin/sshd -D -e; sleep 3; done;
root 4833 0.0 0.0 9888 7600 ? S 08:24 0:00 sshd: /base/mambaforge/bin/sshd -D -e [listener] 0 of 10-100 startups
root 4839 0.2 0.0 17700160 209192 ? Sl 08:24 0:23 /base/mambaforge/bin/python3.10 /base/mambaforge/bin/jupyter-lab --ip 0.0.0.0 --port 8888 --no-browser --ServerApp.
root 5156 0.0 0.0 5124 4280 pts/0 Ss 08:26 0:00 /bin/bash -l
root 5629 0.0 0.0 5124 4248 pts/1 Ss 08:36 0:00 /bin/bash -l
root 7482 0.0 0.0 232328 36616 pts/0 Sl+ 11:02 0:00 uv run scripts/batch_sweep.sh --use-wandb --train-data data/tinystories_train.bin --val-data data/tinystories_val.b
root 7485 0.0 0.0 4892 3612 pts/0 S+ 11:02 0:00 bash scripts/batch_sweep.sh --use-wandb --train-data data/tinystories_train.bin --val-data data/tinystories_val.bin
root 7512 0.0 0.0 232332 36100 pts/0 Sl+ 11:02 0:00 uv run python -m cs336_basics.train --train-data data/tinystories_train.bin --data-dtype uint16 --vocab-size 10000
root 7515 98.0 0.2 31912736 2677372 pts/0 Sl+ 11:02 23:36 /root/shared-nvme/cs336/.venv/bin/python3 -m cs336_basics.train --train-data data/tinystories_train.bin --data-dtyp
root 7651 0.2 0.0 1270524 45484 ? Ssl 11:03 0:03 /root/shared-nvme/cs336/.venv/lib/python3.12/site-packages/wandb/bin/wandb-core --port-filename /tmp/tmpucm1x3jq/po
root 7671 0.0 0.0 1180028 24544 ? Sl 11:03 0:00 /root/shared-nvme/cs336/.venv/lib/python3.12/site-packages/wandb/bin/gpu_stats --portfile /tmp/wandb-system-monitor
root 7782 0.0 0.0 12164 9020 ? Ss 11:09 0:00 sshd-session: root [priv]
root 7784 0.0 0.0 12424 6496 ? S 11:09 0:00 sshd-session: root@notty
root 7785 0.2 0.0 2844 2048 ? Ss 11:09 0:02 /base/mambaforge/libexec/sftp-server
root 8008 0.0 0.0 8468 4276 pts/1 R+ 11:27 0:00 ps aux
nvidia-smi
仅适用于使用nvidia显卡的设备。
可以查看GPU情况，包括CUDA版本、有几个设备和设备的使用率等。
使用效果：
这里是单卡5090，CUDA版本为12.8，因此需要安装2.7.1等PyTorch版本。
(cs336) root@p-4e088649b993-ackcs-00gjgdno:~# nvidia-smi
Tue Mar 10 11:18:13 2026
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 570.172.08 Driver Version: 570.172.08 CUDA Version: 12.8 |
|-----------------------------------------+------------------------+----------------------+
| GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|=========================================+========================+======================|
| 0 NVIDIA GeForce RTX 5090 On | 00000000:D8:00.0 Off | N/A |
| 0% 42C P1 133W / 575W | 1114MiB / 32607MiB | 24% Default |
| | | N/A |
+-----------------------------------------+------------------------+----------------------+
+-----------------------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=========================================================================================|
| 0 N/A N/A 7515 C ...-nvme/cs336/.venv/bin/python3 1104MiB |
+-----------------------------------------------------------------------------------------+
-v -vv -vvv
严格来说这个不是指令，而是一个常见的命令参数。但好用。
有许多工具默认状态给你提供的信息是很有限的，比如 pip 。如果加了 -v ，它就会给你提供更多的信息；-vv 更多更多……
比如你怀疑某个指令是不是卡在网络连接问题上了、又没有信息，就可以加这个参数看看。
使用效果（不加参数）：
(pymc) PS C:\Users\10pi> pip install einops
Collecting einops
Using cached einops-0.8.1-py3-none-any.whl.metadata (13 kB)
Using cached einops-0.8.1-py3-none-any.whl (64 kB)
Installing collected packages: einops
Successfully installed einops-0.8.1
使用效果（-v）：
(pymc) PS C:\Users\10pi> pip install -v einops
Using pip 24.3.1 from C:\Software\Miniconda\envs\pymc\lib\site-packages\pip (python 3.8)
Link requires a different Python (3.8.0 not in: '>=3.9'): https://files.pythonhosted.org/packages/2a/09/f8d8f8f31e4483c10a906437b4ce31bdf3d6d417b73fe33f1a8b59e34228/einops-0.8.2-py3-none-any.whl (from https://pypi.org/simple/einops/) (requires-python:>=3.9)
Link requires a different Python (3.8.0 not in: '>=3.9'): https://files.pythonhosted.org/packages/2c/77/850bef8d72ffb9219f0b1aac23fbc1bf7d038ee6ea666f331fa273031aa2/einops-0.8.2.tar.gz (from https://pypi.org/simple/einops/) (requires-python:>=3.9)
Collecting einops
Obtaining dependency information for einops from https://files.pythonhosted.org/packages/87/62/9773de14fe6c45c23649e98b83231fffd7b9892b6cf863251dc2afa73643/einops-0.8.1-py3-none-any.whl.metadata
Using cached einops-0.8.1-py3-none-any.whl.metadata (13 kB)
Using cached einops-0.8.1-py3-none-any.whl (64 kB)
Installing collected packages: einops
Successfully installed einops-0.8.1
使用效果（-vv）：（我省略了一部分内容）
(pymc) PS C:\Users\10pi> pip install einops -vv
Using pip 24.3.1 from C:\Software\Miniconda\envs\pymc\lib\site-packages\pip (python 3.8)
Non-user install because site-packages writeable
Disabling truststore because Python version isn't 3.10+
Created temporary directory: C:\Users\10pi\AppData\Local\Temp\pip-build-tracker-hni59twj
Initialized build tracking at C:\Users\10pi\AppData\Local\Temp\pip-build-tracker-hni59twj
Created build tracker: C:\Users\10pi\AppData\Local\Temp\pip-build-tracker-hni59twj
Entered build tracker: C:\Users\10pi\AppData\Local\Temp\pip-build-tracker-hni59twj
Created temporary directory: C:\Users\10pi\AppData\Local\Temp\pip-install-8kirg154
Created temporary directory: C:\Users\10pi\AppData\Local\Temp\pip-ephem-wheel-cache-zt8dwlfm
1 location(s) to search for versions of einops:
* https://pypi.org/simple/einops/
Fetching project page and analyzing links: https://pypi.org/simple/einops/
Getting page https://pypi.org/simple/einops/
Found index url https://pypi.org/simple/
Looking up "https://pypi.org/simple/einops/" in the cache
Request header has "max_age" as 0, cache bypassed
Starting new HTTPS connection (1): pypi.org:443
https://pypi.org:443 "GET /simple/einops/ HTTP/1.1" 304 0
Fetched page https://pypi.org/simple/einops/ as application/vnd.pypi.simple.v1+json
Found link https://files.pythonhosted.org/packages/36/1f/a5e6496a167b6e892123b201ccce251184457ce7a1b6c8e06662c09bfbab/einops-0.1.0-py3-none-any.whl (from https://pypi.org/simple/einops/), version: 0.1.0
Found link ...
https://files.pythonhosted.org/packages/2c/77/850bef8d72ffb9219f0b1aac23fbc1bf7d038ee6ea666f331fa273031aa2/einops-0.8.2.tar.gz (from https://pypi.org/simple/einops/) (requires-python:>=3.9)
Skipping link: 0.8.2 Requires-Python >=3.9: https://files.pythonhosted.org/packages/2c/77/850bef8d72ffb9219f0b1aac23fbc1bf7d038ee6ea666f331fa273031aa2/einops-0.8.2.tar.gz (from https://pypi.org/simple/einops/) (requires-python:>=3.9)
Skipping link: not a file: https://pypi.org/simple/einops/
Given no hashes to check 26 links for project 'einops': discarding no candidates
Collecting einops
Obtaining dependency information for einops from https://files.pythonhosted.org/packages/87/62/9773de14fe6c45c23649e98b83231fffd7b9892b6cf863251dc2afa73643/einops-0.8.1-py3-none-any.whl.metadata
Created temporary directory: C:\Users\10pi\AppData\Local\Temp\pip-unpack-g4cpg3nk
Looking up "https://files.pythonhosted.org/packages/87/62/9773de14fe6c45c23649e98b83231fffd7b9892b6cf863251dc2afa73643/einops-0.8.1-py3-none-any.whl.metadata" in the cache
Current age based on date: 19087633
Ignoring unknown cache-control directive: immutable
Freshness lifetime from max-age: 365000000
The response is "fresh", returning cached response
365000000 > 19087633
Using cached einops-0.8.1-py3-none-any.whl.metadata (13 kB)
Created temporary directory: C:\Users\10pi\AppData\Local\Temp\pip-unpack-nq9eoxgz
Looking up "https://files.pythonhosted.org/packages/87/62/9773de14fe6c45c23649e98b83231fffd7b9892b6cf863251dc2afa73643/einops-0.8.1-py3-none-any.whl" in the cache
Current age based on date: 19087579
Ignoring unknown cache-control directive: immutable
Freshness lifetime from max-age: 365000000
The response is "fresh", returning cached response
365000000 > 19087579
Using cached einops-0.8.1-py3-none-any.whl (64 kB)
Downloading link https://files.pythonhosted.org/packages/87/62/9773de14fe6c45c23649e98b83231fffd7b9892b6cf863251dc2afa73643/einops-0.8.1-py3-none-any.whl (from https://pypi.org/simple/einops/) (requires-python:>=3.8) to C:\Users\10pi\AppData\Local\Temp\pip-unpack-nq9eoxgz\einops-0.8.1-py3-none-any.whl
Installing collected packages: einops
Successfully installed einops-0.8.1
Disabling truststore because Python version isn't 3.10+
Remote version of pip: 25.0.1
Local version of pip: 24.3.1
Was pip installed by pip? False
Removed build tracker: 'C:\\Users\\10pi\\AppData\\Local\\Temp\\pip-build-tracker-hni59twj'
-vvv 就不说了。肯定更多。