serhii.net

English

Reading “German: An Essential Grammar” by Donaldson found this bit: ¹

English has a rule that if the time of an event that
occurred in the past is mentioned, then the imperfect must be used, but if
the time is omitted, the perfect is required, e.g. \

• He returned from Hamburg yesterday.
• He has returned from Hamburg.
• He has returned from Hamburg yesterday. (not grammatical)

TIL.

zsh detach and disown

zsh-specific - to detach & disown a process, there’s &!: ²

dolphin &!

German / Deutsch

Long question and answer about fahren zu/nach/in/…: Richtungen und Ziele

German FSI language courses

The Yojik Website has the FSI courses FSI Languages Courses and the website as I remember it.

Taskwarrior

Changed ~/.taskrc to show any active tasks regardless of anything else in my sprint view:

s () {task s \(project:w or \(sprint:$SPRINT \(+A or +O\)\) or +ACTIVE\) "$*"}

Turn off screen/monitor with xset

Standard lock command leaves both monitors on.

Reddit³ mentioned two commands:

xset s activate
xset dpms force off

The second one worked for me!

Now I have shiny new screen lock (and suspend too, while we are at it) keybinding in i3 config!

bindsym $ms+n exec gnome-screensaver-command -l && xset dpms force off
bindsym $ms+Shift+n exec i3lock -i ~/s/black_lock.png -t -p win -e && systemctl suspend -i

Nvidia Docker images

Nvidia has a repo of all docker images it creates, one of them: Torch | NVIDIA NGC

German

“Das finde ich zielführender als…” - heard at work

Docker - automatically assign a free port

docker run --name frontend -p 0:80 frontend:latest¹

Port 0 gets passed to the kernel that assigns any free port.

To see which one, docker port somecontainer.

Docker run container on specific GPU

docker run --gpus device=3 -e NVIDIA_VISIBLE_DEVICES=0 -e CUDA_VISIBLE_DEVICES=0 myservice

Where the device=3 is the GPU id on the host that we want to use.

grep ignore case

lspci | grep -i "nvidia"

-i == ‘ignore case’ is actually something that I can remember.

Docker (stop) autostart of container

Docker will autostart any container with a RestartPolicy of ‘always’ when the docker service initially starts. ¹

I can set/unset it in kitematic, or through terminal:

docker update --restart=no my-container

apt-get purge remove –autoremove etc

Quoting SO: ²

    apt purge --auto-remove <packagename>

purges packagename and any packages which are rendered unnecessary by its removal, as well as any other packages which aren’t necessary.

    apt autoremove --purge

purges any packages which aren’t necessary (marked as “automatically installed” and with no dependent packages).

The first form is what you’d use when manipulating individual packages; the latter is a clean-up operation across all packages.

Ways to clean up with apt-get - tutorial

This seems nice, TODO: Cleaning up with apt-get | Network World

Backing up LVM disk encryption keys

LVM - Debian Wiki is nice and readable. I used this command to backup the headers:

 sudo cryptsetup luksHeaderBackup /dev/nvmeXXXXX   --header-backup-file headerBackupFile

… and put it somewhere not on the drive I’ll be recovering if it all goes wrong.

Setting up Tensorflow and CUDA with an eGPU

Aaaand the saga continues!

…since the GPU is an eGPU, apparently I do need to do the harder way: Accelerating Machine Learning on a Linux Laptop with an External GPU | NVIDIA Developer Blog

Getting the eGPU detected

It is, I can see it:

(17:42:42/10815)~/$ lspci | grep -i VGA
00:02.0 VGA compatible controller: Intel Corporation UHD Graphics 620 (rev 07)
0c:00.0 VGA compatible controller: NVIDIA Corporation GP104 [GeForce GTX 1070] (rev a1)

but if it wasn’t, I’d authorize it and check with boltctl list:

(17:43:13/10817)~/$ boltctl list
[...]
 ● GIGABYTE GV-N1070IXEB-8GD
   ├─ type:          peripheral
   ├─ name:          GV-N1070IXEB-8GD
   ├─ vendor:        GIGABYTE
   ├─ uuid:          # redacted
   ├─ status:        authorized
   │  ├─ domain:     domain0
   │  └─ authflags:  none
   ├─ authorized:    Do 29 Apr 2021 07:57:37 UTC
   ├─ connected:     Do 29 Apr 2021 07:57:37 UTC
   └─ stored:        no

How to setup an eGPU on Ubuntu for TensorFlow describes other things that can go wrong:

I had to disable the following, otherwise my eGPU was not detected:

• Secure Boot
• Thunderbolt Security Level

From this point on, I follow Nvidia’s tutorial ³ unless stated otherwise.

Purging, cleaning up old broken install attempts, updating and upgrading

Using quotes means the * doesn’t have to be escaped.

sudo apt-get purge "nvidia*"

This is a fuller example: ⁴

sudo rm /etc/apt/sources.list.d/cuda*
sudo apt remove --autoremove nvidia-cuda-toolkit
sudo apt remove --autoremove nvidia-*

Found and manually removed /etc/apt/sources.list.d/graphics-drivers-ubuntu-ppa-bionic.list, leaving the .save file in place.

As per nvidia’s guide,

sudo apt-get update
sudo apt-get dist-upgrade

To be safe, rebooted.

Downloading the correct drivers

The existing driver is most likely Nouveau, an open-source driver for NVIDIA GPUs. Because Nouveau doesn’t support eGPU setups, install the NVIDIA CUDA and NVIDIA drivers instead. You must also stop the kernel from loading Nouveau. ³

okay!

Change of plan - what is NVIDIA data-science-stack?

Found this: NVIDIA/data-science-stack: NVIDIA Data Science stack tools Read about it here: Ubuntu for machine learning with NVIDIA RAPIDS in 10 min | Ubuntu

Official by nvidia, and seems to do automatically what’s needed for supported systems. Let’s run a script from the internet that installs drivers, loads kernel modules etc.

Source is available, yay for open source: data-science-stack/data-science-stack at master · NVIDIA/data-science-stack

Ran ./data-science-stack setup-system - uses sudo, didn’t ask for root or anything.o

Seems to have installed nvidia driver version 460. Asked to reboot at the end.

Rebooted.

(18:40:30/10909)~/$ nvidia-smi
NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

okay. Same results I had. Confirms that my prev. steps weren’t wronger than the script.

(18:41:49/10910)~/$ sudo apt list --installed | grep "\(cuda\|nvidia\)"

WARNING: apt does not have a stable CLI interface. Use with caution in scripts.

libnccl2/unknown,now 2.9.6-1+cuda11.3 amd64 [installed]
libnvidia-cfg1-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-common-460/unknown,now 460.73.01-0ubuntu1 all [installed,automatic]
libnvidia-compute-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-container-tools/bionic,now 1.4.0-1 amd64 [installed,automatic]
libnvidia-container1/bionic,now 1.4.0-1 amd64 [installed,automatic]
libnvidia-decode-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-encode-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-extra-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-fbc1-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-gl-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
libnvidia-ifr1-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
nvidia-compute-utils-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
nvidia-container-runtime/bionic,now 3.5.0-1 amd64 [installed,automatic]
nvidia-container-toolkit/bionic,now 1.5.0-1 amd64 [installed,automatic]
nvidia-dkms-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
nvidia-docker2/bionic,now 2.6.0-1 all [installed]
nvidia-driver-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed]
nvidia-kernel-common-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
nvidia-kernel-source-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
nvidia-prime/bionic-updates,bionic-updates,now 0.8.16~0.18.04.1 all [installed,automatic]
nvidia-settings/unknown,unknown,now 465.19.01-0ubuntu1 amd64 [installed,automatic]
nvidia-utils-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]
xserver-xorg-video-nvidia-460/unknown,now 460.73.01-0ubuntu1 amd64 [installed,automatic]

Also, as usual,

(18:48:34/10919)~/$ lsmod | grep nvi
(18:48:37/10920)~/$

lspci -k shows the kernel modules:

0c:00.0 VGA compatible controller: NVIDIA Corporation GP104 [GeForce GTX 1070] (rev a1)
        Subsystem: Gigabyte Technology Co., Ltd GP104 [GeForce GTX 1070]
        Kernel modules: nvidiafb, nouveau

This output implies no nvidia driver is installed on my system⁵. …though it is.

$ nvidia-settings --version
nvidia-settings:  version 465.19.01

software-properties-gtk tells me I’m using the proprietary nvidia-driver-460, not 465

In any case, can’t blacklist nouveau as still there are no ubuntu kernel modules.

BUT!

(19:04:04/10946)~/$ dkms status
nvidia, 460.73.01: added

Also, inxi -Fxxxrz (found somewhere on the internet):

Graphics:  Card-1: Intel UHD Graphics 620 bus-ID: 00:02.0 chip-ID: 8086:5917
           Card-2: NVIDIA GP104 [GeForce GTX 1070] bus-ID: 0c:00.0 chip-ID: 10de:1b81
           Display Server: x11 (X.Org 1.19.6 ) drivers: modesetting,nvidia (unloaded: fbdev,vesa,nouveau)

It it sees them as there and loaded? Does dkms somehow bypass lsmod etc?

sudo dkms autoinstall should autoinstall all added drivers, …let’s hope for the best I guess.

(19:11:47/10958)~/$ sudo dkms autoinstall

Kernel preparation unnecessary for this kernel.  Skipping...
applying patch disable_fstack-clash-protection_fcf-protection.patch...patching file Kbuild
Hunk #1 succeeded at 85 (offset 14 lines).


Building module:
cleaning build area...
unset ARCH; [ ! -h /usr/bin/cc ] && export CC=/usr/bin/gcc; env NV_VERBOSE=1 'make' -j8 NV_EXCLUDE_BUILD_MODULES='' KERNEL_UNAME=5.4.0-72-generic IGNORE_XEN_PRESENCE=1 IGNORE_CC_MISMATCH=1 SYSSRC=/lib/modules/5.4.0-72-generic/build LD=/usr/bin/ld.bfd modules......(bad exit status: 2)
ERROR: Cannot create report: [Errno 17] File exists: '/var/crash/nvidia-dkms-460.0.crash'
Error! Bad return status for module build on kernel: 5.4.0-72-generic (x86_64)
Consult /var/lib/dkms/nvidia/460.73.01/build/make.log for more information.

The file is long, keys seems:

 scripts/Makefile.build:269: recipe for target '/var/lib/dkms/nvidia/460.73.01/build/nvidia/nv.o' failed
 make[2]: *** [/var/lib/dkms/nvidia/460.73.01/build/nvidia/nv.o] Error 1
 Makefile:1754: recipe for target '/var/lib/dkms/nvidia/460.73.01/build' failed
 make[1]: *** [/var/lib/dkms/nvidia/460.73.01/build] Error 2
 make[1]: Leaving directory '/usr/src/linux-headers-5.4.0-72-generic'
 Makefile:80: recipe for target 'modules' failed
 make: *** [modules] Error 2
DKMSKernelVersion: 5.4.0-72-generic
Date: Fri Apr 30 18:30:45 2021
DuplicateSignature: dkms:nvidia-dkms-460:460.73.01-0ubuntu1:/var/lib/dkms/nvidia/460.73.01/build/conftest/functions.h:11:2: error: #error acpi_walk_namespace() conftest failed!
Package: nvidia-dkms-460 460.73.01-0ubuntu1
PackageVersion: 460.73.01-0ubuntu1
SourcePackage: nvidia-graphics-drivers-460
Title: nvidia-dkms-460 460.73.01-0ubuntu1: nvidia kernel module failed to build

Smells like a driver/kernel support isse?

First result when googling dkms nvidia 460 is this: Can’t get nvidia 460 module to build on Ubuntu 20.04 to support two A100s - GPU Unix Graphics / Linux - NVIDIA Developer Forums

Please check if the build symlink to the headers for dkms exists:

ls /lib/modules/$(uname -r)/build

Otherwise, create it

ln -s /usr/src/linux-headers-$(uname -r)  /lib/modules/$(uname -r)/build

Didn’t have it, created it, trying again, same error, deleted the previous log, full output is:

(19:19:54/10967)~/$ sudo dkms autoinstall

Kernel preparation unnecessary for this kernel.  Skipping...
applying patch disable_fstack-clash-protection_fcf-protection.patch...patching file Kbuild
Hunk #1 succeeded at 85 (offset 14 lines).


Building module:
cleaning build area...
unset ARCH; [ ! -h /usr/bin/cc ] && export CC=/usr/bin/gcc; env NV_VERBOSE=1 'make' -j8 NV_EXCLUDE_BUILD_MODULES='' KERNEL_UNAME=5.4.0-72-generic IGNORE_XEN_PRESENCE=1 IGNORE_CC_MISMATCH=1 SYSSRC=/lib/modules/5.4.0-72-generic/build LD=/usr/bin/ld.bfd modules.......(bad exit status: 2)
Error! Bad return status for module build on kernel: 5.4.0-72-generic (x86_64)
Consult /var/lib/dkms/nvidia/460.73.01/build/make.log for more information.

The file is full of what looks like syntax errors..?

This charming chinese website seems to imply gcc version is to blame: NVIDIA驱动出错：NVIDIA-SMI has failed because it couldn‘t communicate with the NVIDIA driver. Make sure t_sazass的博客-CSDN博客

(19:22:39/10974)~/$ cat /proc/version
Linux version 5.4.0-72-generic (buildd@lgw01-amd64-021) (gcc version 7.5.0 (Ubuntu 7.5.0-3ubuntu1~18.04)) #80~18.04.1-Ubuntu SMP Mon Apr 12 23:26:25 UTC 2021

sudo apt install gcc-8
sudo update-alternatives --config gcc
sudo update-alternatives --remove-all gcc
sudo update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-8 10
sudo update-alternatives --install /usr/bin/cc cc /usr/bin/gcc-8 10

Let’s retry dkms autoinstall:

(19:26:03/10981)~/$ sudo dkms autoinstall

Kernel preparation unnecessary for this kernel.  Skipping...
applying patch disable_fstack-clash-protection_fcf-protection.patch...patching file Kbuild
Hunk #1 succeeded at 85 (offset 14 lines).


Building module:
cleaning build area...
unset ARCH; [ ! -h /usr/bin/cc ] && export CC=/usr/bin/gcc; env NV_VERBOSE=1 'make' -j8 NV_EXCLUDE_BUILD_MODULES='' KERNEL_UNAME=5.4.0-72-generic IGNORE_XEN_PRESENCE=1 IGNORE_CC_MISMATCH=1 SYSSRC=/lib/modules/5.4.0-72-generic/build LD=/usr/bin/ld.bfd modules...............
Signing module:
 - /var/lib/dkms/nvidia/460.73.01/5.4.0-72-generic/x86_64/module/nvidia-modeset.ko
 - /var/lib/dkms/nvidia/460.73.01/5.4.0-72-generic/x86_64/module/nvidia.ko
 - /var/lib/dkms/nvidia/460.73.01/5.4.0-72-generic/x86_64/module/nvidia-uvm.ko
 - /var/lib/dkms/nvidia/460.73.01/5.4.0-72-generic/x86_64/module/nvidia-drm.ko
Secure Boot not enabled on this system.
cleaning build area...

DKMS: build completed.

nvidia.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.4.0-72-generic/updates/dkms/

nvidia-modeset.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.4.0-72-generic/updates/dkms/

nvidia-drm.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.4.0-72-generic/updates/dkms/

nvidia-uvm.ko:
Running module version sanity check.
 - Original module
   - No original module exists within this kernel
 - Installation
   - Installing to /lib/modules/5.4.0-72-generic/updates/dkms/

depmod...

DKMS: install completed.

WOW. WOOOOOW. WOOOOOOOOOOOOOOOOOOOOOO

Without even restarting, after the first command my screen flashed and changed resolution a bit, BUT THEN IT WORKED

(19:34:17/10983)~/$ nvidia-smi
No devices were found
(19:34:20/10984)~/$ nvidia-smi
Fri Apr 30 19:34:22 2021
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.73.01    Driver Version: 460.73.01    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+
| 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  GeForce GTX 1070    On   | 00000000:0C:00.0 Off |                  N/A |
|  0%   54C    P0    37W / 151W |      7MiB /  8119MiB |      0%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

All these attempts failed because the nvidia module in dkms couldn’t install because syntax errors because old gcc compiler version.

What could I have done differently? Why at no point did I see errors about the kernel module failing to build, where should I have looked for them? And why syntax errors instead of something checking the used gcc version and loudly failing when there was a mismatch? Why is that chinese website the only place I found this fix?

(19:42:57/10995)~/$ lsmod | grep nvidia
nvidia_uvm           1015808  0
nvidia_drm             57344  1
nvidia_modeset       1228800  1 nvidia_drm
nvidia              34123776  17 nvidia_uvm,nvidia_modeset
drm_kms_helper        188416  2 nvidia_drm,i915
drm                   491520  15 drm_kms_helper,nvidia_drm,i915

Now let’s hope this survives a restart. And that it works when the eGPU is disconnected.

NVIDIA data-science-stack

Following the readme, ran both options in separate terminals:

./data-science-stack list
./data-science-stack build-container
./data-science-stack run-container

and

./data-science-stack list
./data-science-stack build-conda-env
./data-science-stack run-jupyter

The latter seems to be installing CUDA and friends on my computer - didn’t expect it, but I need them either way I think, I guess I’ll let the script handle everything since it started. It installed conda to ~/conda/, but again, not sure what I was expecting

Both running for 20+ minutes now

EDIT: ~/conda/ took 20gb filling up my drive, blocking everything, deleted it

The docker with jupyterlab - tensorflow can’t access the GPU, but pytorch can.

Carrying on with setting the eGPU up

The NVIDIA eGPU tutorial³ continues with offloading Xorg to the GPU - do I want this? Can I use the GPU just for training, and leave Xorg running on the internal one? I probably don’t

Restarting and testing

As I remember from the last time, X doesn’t start when the GPU is connected at boot but everything’s fine when it gets connected after starting X. When it’s connected, it seems the driver gets loaded and nvidia-smi etc works. That the system works without the eGPU attached is nice! Plug-and-play is nice too.

Installed pytorch in a virtualenv, for cuda 11.1, test snippet says cuda works!

import torch
x = torch.rand(5, 3)
print(x)

torch.cuda.is_available()

Tensorflow:

>>> import tensorflow as tf
2021-04-30 21:36:12.984883: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
>>> tf.debugging.set_log_device_placement(True)
>>> a = tf.constant([[1.0, 2.0, 3.0], [4.0, 5.0, 6.0]])
2021-04-30 21:36:23.055614: I tensorflow/compiler/jit/xla_cpu_device.cc:41] Not creating XLA devices, tf_xla_enable_xla_devices not set
2021-04-30 21:36:23.058062: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcuda.so.1
2021-04-30 21:36:23.115366: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:941] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-04-30 21:36:23.116510: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1720] Found device 0 with properties:
pciBusID: 0000:0c:00.0 name: GeForce GTX 1070 computeCapability: 6.1
coreClock: 1.721GHz coreCount: 15 deviceMemorySize: 7.93GiB deviceMemoryBandwidth: 238.66GiB/s
2021-04-30 21:36:23.116553: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcudart.so.11.0
2021-04-30 21:36:23.119974: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublas.so.11
2021-04-30 21:36:23.120034: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcublasLt.so.11
2021-04-30 21:36:23.121503: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcufft.so.10
2021-04-30 21:36:23.121842: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcurand.so.10
2021-04-30 21:36:23.125037: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusolver.so.10
2021-04-30 21:36:23.125803: I tensorflow/stream_executor/platform/default/dso_loader.cc:49] Successfully opened dynamic library libcusparse.so.11
2021-04-30 21:36:23.125980: W tensorflow/stream_executor/platform/default/dso_loader.cc:60] Could not load dynamic library 'libcudnn.so.8'; dlerror: libcudnn.so.8: cannot open shared object file: No such file or directory
2021-04-30 21:36:23.125996: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1757] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.

Which libcudnn?

Tensorflow’s tutorial (GPU support  |  TensorFlow) does this:

Install development and runtime libraries (~4GB)
sudo apt-get install --no-install-recommends \
    cuda-11-0 \
    libcudnn8=8.0.4.30-1+cuda11.0  \
    libcudnn8-dev=8.0.4.30-1+cuda11.0

What is the version for CUDA 11.2? cuDNN Archive | NVIDIA Developer has download links. The one for 11.2 is called “cudnn-11.2-linux-x64-v8.1.1.33.tgz”. I plug those versions in, they exist and install fine:

sudo apt-get install   libcudnn8=8.1.1.33-1+cuda11.2
sudo apt-get install   libcudnn8-dev=8.1.1.33-1+cuda11.2

And tensorflow now works!

2021-04-30 21:42:46.176942: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1406] Created TensorFlow device (/job:localhost/replica:0/task:0/device:GPU:0 with 7440 MB memory) -> physical GPU (device: 0, name: GeForce GTX 1070, pci bus id: 0000:0c:00.0, compute capability: 6.1)

I can’t believe it but wow. It’s finished, it works, X didn’t die, plug-and-play works, no manual driver loading.

All in all, including all the failed attempts, took 5:30h of pure time, according to my time tracking.

The only wrinkle is that X doesn’t start when turning the computer on with the eGPU attached, but I can 100% live with that!

GPU benchmarking linux

How to Benchmark your GPU on Linux has a fun quote:

This tool is very old, very basic and only tests a small portion of today’s OpenGL capabilities. Back in the old days, it was used to determine if the proprietary driver was installed and running properly as open-source drivers were performing awfully enough to be perfectly noticeable during this test. Nowadays, you won’t notice any difference between the two

qutebrowser open a private window

Added this to config.py:

config.bind('<Alt-P>', 'set-cmd-text -s :open -p ')

Managing dotfiles with machine-specific configuration

Qutebrowser import other config files

Seen in someone’s config.py on gitlab⁶:

for f in glob.glob(str(config.configdir / 'conf.d/*.py')):
    config.source(str(os.path.relpath(f, start=config.configdir)))

Random i3 configs

Nice examples: i3_config/settings.d at master · kiddico/i3_config · GitHub

i3 doesn’t have any kind of include directive in the config files, sadly. i3 - Source/import file from i3wm config - Stack Overflow is one option:

bindsym $mod+Shift+c exec "cat ~/.config/i3/colors ~/.config/i3/base > ~/.config/i3/config && i3-msg reload"

A keybinding to overwrite the config file and restart i3 with a command.

To read - life hacking

This looks very interesting, I shouldn’t forget to go through this: Life Hacking His blog with personal examples: Alex Vermeer — Life-Hacking. Climbing. Striving for awesome. Coffee. — Page 2

A non-pdf description of Life Areas with questions and metrics for each.

(He’s the same guy who created the awesome How to Get Motivated: A Guide for Defeating Procrastination poster!)

And let’s remember the classic: Evidence-based advice on how to be successful in any job - 80,000 Hours

Detach process completely from terminal

Two options I like:⁷

• nohup cmd &
• cmd & disown

I feel one of these will become part of many aliases of mine.

And short bash function from the same place:

function dos() {
    # run_disowned and silenced

    run_disowned "$@" 1>/dev/null 2>/dev/null
}

PEP8

To read: PEP 8 – Style Guide for Python Code | Python.org

English / random

• “If you feel a misalignment with …”
• Ticketize (verb)

Jira ticket search and filtering

I should learn about the search syntax for jira tickets:

assignee = currentuser() and statusCategory != Done ORDER BY updated DESC

Installing CUDA and pytorch and tensorflow

Following this: CUDA 10.1 installation on Ubuntu 18.04 LTS | Medium nope, errors

In the same github discussion about installing CUDA on ubuntu that I’ve been to twice this bit is mentioned: ¹

The very very important thing is that never install “nvidia-driver-***” driver by yourself.

Required nvidia drivers are installed while doing sudo apt install -y cuda=10.0.130-1

Zsh wildcards and apt-get remove

sudo apt remove --autoremove nvidia-* doesn’t work as-is in zsh! * gets interpreted as files in current directory. Explains my CUDA issues, everything seemed to work till I ran the above in a directory containing files with matching names that got helpfully shown.

sudo apt remove --autoremove nvidia-\* is the answer.

(or 'nvidia-*')

Not the first time this bites me, at least the third, and all of them in the context of doing CUDA stuff.

German

“Es funktioniert fabelhaft” - heard at work

Purging packages

apt --fix-broken install didn’t help as advertised, but removing all the broken packages together with sudo dpkg -P cuda-libraries-10-0 libnvidia-common-390 helped! After this removing/cleaning up everything else worked. A lot of this mentioned changes to initramfs, I really hope I’ll be able to boot up next time :(

Also - if 90% of the tutorials about how to install $thing start with “Remove any traces of installs of $thing you have” it’s a nice sign that something’s shady.

Docker logs

docker logs 09348209840239

i3 skype floating window fix

Skype fix : i3wm:

Option 1: hide the floating window:

for_window [title="^Skype$" floating] move scratchpad

Option 2:

Clever idea. Although, are you talking about the little window that can be disabled in Skype’s “Settings > Calling > Show call window when Skype is in the background”?

Slack show all messages in all channels

In search, before:Tomorrow is a nice catch-all filter

Pytorch installs its own CUDA!

Your system installations of CUDA and cudnn won’t be used, if you install PyTorch binaries with these libs. E.g. conda install pytorch torchvision cudatoolkit=10.1 -c pytorch will install CUDA 10.1 and cudnn in your current conda environment. ²

Tensorflow CUDA Docker doesn’t need CUDA on host machine, only the nvidia drivers

Nvidia drivers are needed on host machine, but not CUDA! ³

Random / UX / Design?

On TF’s official CUDA install page⁴, the bash listings (that are usually copypasted) contain the standard $ at the beginning, it’s visible, but not copypastable!

Installing CUDA 11.0 using official Tensorflow tutorial

So, hopefully the last time today, as the previous couple of times I end up in the official TF tutorial⁴ about installing CUDA. Armed with the knowledge that:

• pytorch installs its own CUDA and doesn’t care, as long as GPU drivers are there
• Docker installs its own CUDA and doesn’t care, as long as GPU drivers are on the host machine
• Installing nvidia drivers should not be manual, it has to be done by the cuda packages

Snippet:

# Add NVIDIA package repositories
wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-ubuntu1804.pin
sudo mv cuda-ubuntu1804.pin /etc/apt/preferences.d/cuda-repository-pin-600
sudo apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pub
sudo add-apt-repository "deb https://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/ /"
sudo apt-get update

wget http://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb

sudo apt install ./nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.deb
sudo apt-get update

wget https://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/libnvinfer7_7.1.3-1+cuda11.0_amd64.deb
sudo apt install ./libnvinfer7_7.1.3-1+cuda11.0_amd64.deb
sudo apt-get update

# Install development and runtime libraries (~4GB)
sudo apt-get install --no-install-recommends \
    cuda-11-0 \
    libcudnn8=8.0.4.30-1+cuda11.0  \
    libcudnn8-dev=8.0.4.30-1+cuda11.0

# Reboot. Check that GPUs are visible using the command: nvidia-smi

# Install TensorRT. Requires that libcudnn8 is installed above.
sudo apt-get install -y --no-install-recommends libnvinfer7=7.1.3-1+cuda11.0 \
    libnvinfer-dev=7.1.3-1+cuda11.0 \
    libnvinfer-plugin7=7.1.3-1+cuda11.0

Done, no conflicts, no anything, worked better than most Medium tutorials I’ve read today.

# Reboot.

Let’s hope for the best.

UPD: no black screen, booted fine, but nvidia-smi sees no driver.

sudo apt list --installed shows all cuda stuff and nvidia driver to be installed:

nvidia-driver-465/unknown,unknown,now 465.19.01-0ubuntu1 amd64 [installed,automatic]

More worryingly, I see mentions of cuda-10-1 and cuda-11-1 together

list processes ubuntu

I should use ps axf instead of ps aux, the former gives a nice tree representation

Nvidia CUDA official installer documentation

Yet another place that makes it look easy: CUDA Toolkit 11.0 Download | NVIDIA Developer

Docker stuff

• Making it run as non-root: Post-installation steps for Linux | Docker Documentation
  • newgrp docker has to be run from each cli you’ll be using docker from?.. Until you restart
• Best tutorial ever can be started with: docker run -d -p 80:80 docker/getting-started
  • It will start as docker image
  • Very readable and step-by-step
• Docker compose
  • Get started with Docker Compose | Docker Documentation
• Random docker stop accepts the full name (distracted_perlman), but part of its container_id works!
• Unintuitively, the COPY instruction from a Dockerfile copies the contents of the directory, but not the directory itself! ¹

Clean up journalctl

Logs take space (4gb on my box!). To see how much specifically journalctl does:²

journalctl --disk-usage
sudo journalctl --vacuum-time=3d

Jupyter notebooks has terminals!

New -> Terminal. (Which you can use to access your docker running jupyter-notebook)

Docker build contexts and relative paths

$ docker build -t dt2test -f ./docker/Dockerfile . - passes the Dockerfile as explicit parameter, inside it paths are relative to the folder you run docker build in.

For docker compose:

#docker-compose.yml
version: '3.3'    
services:
      yourservice:
        build:
          context: ./
          dockerfile: ./docker/yourservice/Dockerfile

A lot of other nice options at Docker: adding a file from a parent directory - Stack Overflow

Python dataclasses

• This module provides a decorator and functions for automatically adding generated special methods such as init() and repr() to user-defined classes. It

• dataclasses — Data Classes — Python 3.9.4 documentation

HuggingFace

“Token classification” includes but is not limited to NER: Hugging Face – The AI community building the future.. Really nice new correct phrase I’ll be using!

Installing (after tensorflow and/or pytorch):

pip install transformers

Caches by default in user folder but can be overridden:

export HF_HOME="/data/sh/experiments/bert/cache" 

The “hosted inference API” on the website is really cool! dslim/bert-base-NER · Hugging Face

Example of converting conll dataset to what BERT expects: Fine Tuning BERT for NER on CoNLL 2003 dataset with TF 2.0 | by Bhuvana Kundumani | Analytics Vidhya | Medium

The BERT model documentation shows the tokenizers etc etc etc. - BERT — transformers 4.5.0.dev0 documentation

• Training and fine-tuning — transformers 4.5.0.dev0 documentation - same model can be trained/imported from TF to pytorch and back! Wow!

• Documentation of a sample model: transformers/examples/research_projects/distillation at master · huggingface/transformers

  • It has examples of preparing data for finetuning
  • In general, HF’s examples are wonderful

• Another example of fine-tuning BERT in Pytorch for NER: transformers/examples/pytorch/token-classification at master · huggingface/transformers

  • Needs transformers installed from source (git/master): https://huggingface.co/transformers/installation.html#installing-from-source / pip install git+https://github.com/huggingface/transformers
  • Trained in 37 minutes, wrote everything to /tmp/test-ner/, checkpoints, eval data. Wow.
  • Command used was:

  CUDA_VISIBLE_DEVICES=1; python run_ner.py   --model_name_or_path bert-base-uncased   --dataset_name conll2003   --output_dir /tmp/test-ner   --do_train   --do_eval
  

  • I could train it on the Germaner (explore) dataset, it has no validation set and the script doesn’t support automagically splitting using CLI parameters (though the script supports a lot of other flags).

python datasets package

Here datasets is imported: transformers/requirements.txt at master · huggingface/transformers

TODO - what is this and where can I learn more? Is this HF specific? What else is there?

HuggingFace datasets

It has a really nice interface for searching datasets! Filter by task, language, etc.

German NER datasets: Hugging Face – The AI community building the future.

Some German NER models, sometimes based on bert: Hugging Face – The AI community building the future.

• I could try to reproduce them, for example this one using BERT-base-german-cased and finetuned on legal entity recognition: mrm8488/bert-base-german-finetuned-ler · Hugging Face

Huggingface converting between tf and pytorch

Converting Tensorflow Checkpoints — transformers 4.5.0.dev0 documentation

Is this real?

export BERT_BASE_DIR=/path/to/bert/uncased_L-12_H-768_A-12

transformers-cli convert --model_type bert \
  --tf_checkpoint $BERT_BASE_DIR/bert_model.ckpt \
  --config $BERT_BASE_DIR/bert_config.json \
  --pytorch_dump_output $BERT_BASE_DIR/pytorch_model.bin

Random / recipes / cooking

Tatar von geräuchertem Forellenfilet mit Avocado - Annemarie Wildeisens KOCHEN

Die Forellenfilets in kleine Würfelchen schneiden. Die Schalotte schälen und sehr fein hacken. Die Cherrytomaten je in 6 oder 8 Stücke schneiden. Alle diese Zutaten in eine kleine Schüssel geben und sorgfältig mit der Mayonnaise mischen.

Forelle + tomatos + mayonnaise is literally the only recipe I’ve liked with mayonnaise in it

Jira old issue view + qutebrowser config setting

To redirect an issue to the old view, add ?oldIssueView=true.

Added this to config.py:

config.bind('<Ctrl-J>', ':open {url}?oldIssueView=true')

Ubuntu screen apt-get

(18:03:38/10185) sudo apt install screen
# ...
Suggested packages:
  byobu | screenie | iselect
The following NEW packages will be installed:

… did I just get an advert for a competitor when installing screen? :) Since when does ubuntu do this and where can I read more about it?

Deutsch / German

“Meetingtourismus oder Papiergenerieren?” (heard at work)

Qutebrowser userscripts

It seems to run userscripts not in the virtualenv qutebrowser uses, but the standard system one? Installing packages in virtualenv didn’t work, but installing them globally did.

DVC

Moving/renaming a file/directory is easy: dvc move from to¹. Automatically updates the from.dvc files. Then .gitignore and the .dvc file have to be added and committed through git as usual.

Data organization / dataset structure link

This is interesting: Data Organization — documentation

In general: Best Practices for Scientific Data Management — documentation

This guide describes Axiom Data Science’s best practices for scientific data management. The intent of these practices is to improve the accessibility and usability of your data. These practices may be followed at any time during the preparation of your dataset, but are most useful when considered at the onset of project planning and implemented during data collection.

Also related: Organising your data | Research Data Management

Tree output only directories

tree -d does it.

Git paths from root of repo

Root of repo: git rev-parse --show-prefix ²

--git-dir returns the location of the .git folder, and --show-toplevel returns the absolute location of the git root.

Patterns / phrases / Random

• “It’s not a solution, but it’s an approach” - heard at work, VF