Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
BrainDance
May 8, 2007

Disco all night long!

Hey guys, I've ran into a problem with Linux and I've tried about everything that's within my skill level to do. I'm on Mint 18 and have tried kernel 4.4.0, 4.8.0 and 4.9.6 that I built myself (based on the Ubuntu configs because my first two attempts at doing it myself wouldn't boot) with practically everything under HD Audio enabled. I have video out over HDMI but I can not get the HDMI display to appear as an audio device. It works fine in Windows (shows up as "nvidia high definition audio" there). Checked the sound preferences both output and hardware and did aplay -l and it only shows one sound device. I updated my graphics and audio drivers, right now I'm using "oem-audio-hda-daily" from ubuntu-audio-dev/alsa-daily and my gpu drivers are nvidia-378, although I've tried a handful of drivers leading up to there. Tried making sure alsamixer and pulseaudio were installed and it made no difference. Turned off automute in alsamixer no luck, pavucontrol only lists speakers and headphone.

On older kernels dmesg | grep '00:1f.3' showed "snd_hda_intel 0000:00:1f.3: failed to add i915_bpo component master (-19)" but with an updated kernel that doesnt show up, although still no sound.
I tried killing pulseaudio and then seeing if it showed up, and I went through a lot of troubleshooting in the ALSA irc chat. At that time I wasn't on 4.9.6 so these logs might reflect that, but I dont know if it'll make much of a difference since updating the kernel didn't do anything.
lspci -nn returned this http://termbin.com/fmz5
`xrandr returned this http://termbin.com/b4z9
`xrandr --output HDMI-0 --set audio on` returned this http://pastebin.com/xE2FR0bi
The ALSA information script returned this http://www.alsa-project.org/db/?f=f097acbb9fb91ef94ea2a78582b4cfa0096ec9f9
glxinfo | grep OpenGL (which is probably not that useful now since I was on a different kernel and a different gpu driver then) returned this http://pastebin.com/pZjgDhVc

I took a look at this ftp://download.nvidia.com/XFree86/gpu-hdmi-audio-document/gpu-hdmi-audio.html and went through the troubleshooting stuff there, but there wasn't much I could do since my computer isn't even seeing that hdmi exists as an audio device (in Linux)
The hdmi is through a mobile gtx 1070. I'm at a loss, I really need hdmi audio for what I use this computer for, but I just have no idea where to go from here. Any ideas?

Adbot
ADBOT LOVES YOU

BrainDance
May 8, 2007

Disco all night long!

I'm gonna be completely honest here, I have no idea how to even check that :/ sorry.

BrainDance
May 8, 2007

Disco all night long!

Alright, here's what I got http://pastebin.com/q6gA0559

BrainDance
May 8, 2007

Disco all night long!

Ok, http://pastebin.com/vvkUTqnz here's the whole thing.

There are quite a few warnings and errors, but I don't know if they're relevant or if they're just because I'm bad at building a kernel.

BrainDance
May 8, 2007

Disco all night long!

It is a laptop but I'm not sure about switchable graphics, I posted the results from lsmod up a bit but heres a second run just in case anything had changed or something http://pastebin.com/tm2iyiZK

edit: Specifically this laptop http://store.asus.com/us/item/201608AM120000028-ASUS+ROG+GL502VS-DB71+15.6

BrainDance fucked around with this message at 16:41 on Jan 29, 2017

BrainDance
May 8, 2007

Disco all night long!

I think so. Is this right? " sudo su -c 'echo 1 > /sys/bus/pci/rescan'"

BrainDance
May 8, 2007

Disco all night long!

telcoM posted:

You can use multiple acpi_osi boot options if necessary: see linux-<version>/Documentation/kernel-parameters.txt.

I've been trying out different combinations of these (and hopefully doing it right, since it doesn't seem to give me any feedback. Is there a way to force it to be verbose instead of giving me the Mint boot screen? There must be.) But I'm thinking, at this point with everything I've tried maybe my hardware is just too new. Mobile gtx 10x0 cards didn't exist before late August.

After reading something about the 4.10 kernel being mostly drivers I decided "gently caress it, using a pre-release kernel is probably bad in almost every way for someone who doesn't know what they're doing but what have I got to lose at this point" (after already killing everything on the previous install by mistake.) So I managed to get 4.10 rc6 installed, but either the nvidia drivers dont work on it, or I'm not doing the right things to get them to work, so I cant see if that even begins to solve my problem because without the nvidia drivers working I don't even get an hdmi display let alone audio. But by then it was way into the morning and I just had to put it aside for a while. So, I don't know. I really want to make this work, but this is turning into a kind of torture at this point.

BrainDance
May 8, 2007

Disco all night long!

Hey, I have a quick question, but it might be very specific and niche so, sorry. I am probably mis-configuring something, but basically I need to be able to type in English and also simplified Chinese with a pinyin keyboard.

I'm on Mint 18.2 xfce. I installed Fcitx to manage the two input methods. I've also installed the google pinyin package (I tried out a couple other pinyin packages too, but it makes no difference.)

If I switch to pinyin I can start typing Chinese characters almost as expected. The thing is, with Chinese, there are many characters that are the same in pinyin (or just have different tones.) On Windows and other OSs, when you type a word like "shi" for example it will show a little menu with all the possible "shi" characters. I swear at one point on another install Mint was doing this too. Now, I dont get any window showing me the possible characters as I type. I just type, and I guess it guesses the first possible "shi" and goes with that.

I've dug through the fcitx configs and tried tweaking a few things, but I cant find much that seems relevant and nothing I do gets me the Chinese character bar. Any ideas how to make this work?


Here's an example of another OS handling this correctly.

BrainDance
May 8, 2007

Disco all night long!

mystes posted:

This is probably a silly question, but are you sure you're actually using fcitx? The input method settings are always complicated because there are different settings qt, gtk, etc., although hopefully whatever you use to configure it is setting up all of these things.

Are you trying to enter text into a gtk program? A qt program? Or a terminal? If you've only tried one of these you might want to try another just to see if there is a difference.

When you say you installed the input methods, you enabled them in fcitx, right?

fcitx also has to be running, but if you actually can actually use its configuration utility and see the input methods and can input stuff using it, then obviously that isn't the issue.

For japanese at least, fcitx is definitely able to display the candidates properly, so unless there's some issue with the Chinese input methods for it specifically I don't know why that wouldn't be working.

Also, I had some issues with the input window appearing in the wrong place with Wayland, but I don't think Mint uses Wayland so that shouldn't be an issue. (You also probably would have noticed if the candidate window was just appearing in the wrong place such as at the upper left of your screen, I imagine.)

I thought so, xfce (mint?) has a little language settings program that lets you choose input method, and I made sure fcitx was always chosen.

Although I seem to have accidentally fixed it. I purged fcitx-config-gtk, fcitx-frontend-gtk2, fcitx-ui-classic and fcitc-googlepinyin, and then reinstalled all of them. Suddenly more options appear in the same place I was looking before, related to appearance of the text display window. Works now though, don't know why it wasn't working in the first place but whatever.


I have another question though that just came up. It's unrelated to the other stuff, but still Mint, xfce. I have a 2nd display, and I would like xfce4-terminal to run fullscreen, no window borders on the second display at logon. I've been reading through xfce-terminal's man pages and put together something that almost does it. 'xfce4-terminal --display=2 --maximize --hide-borders'. It doesn't start it on my 2nd monitor though (and I've tried it with other numbers too, I have this other program that needs the display specified, and setting it to 3 gets it to output on this other monitor. Not with --display= for xfce4-terminal though)

I've tried also "--default-display=2" but, same deal, just goes fullscreen on my first, main display.
I'm using the proprietary nvidia drivers. In nvidia's x server settings it shows both physical displays but then makes them both a part of "x screen 0", like it was one 3840x1080 display spanning across both monitors. Is that maybe part of the problem? How else could I go about doing this?

BrainDance
May 8, 2007

Disco all night long!

Hey everybody, I'm running Mint 18.3 and I think it just got boned somehow.

I noticed the mint updater icon wasn't in the panel so I thought "ehh, I'll go run it and see what's up". This is what's up;

$ mintupdate
$ Traceback (most recent call last):
File "/usr/lib/linuxmint/mintUpdate/mintUpdate.py", line 15, in <module>
import pycurl
ImportError: pycurl: libcurl link-time version (7.47.0) is older than compile-time version (7.52.1)


I have no idea how (or exactly what) happened. The only thing I changed at all today was adding the pipelight package from ppa:pipelight/stable, and I installed filezilla and did normal updates. I'm looking at that stuff now though and it doesn't depend on, recommend or provide any packages related to curl or python. If it's not that, what could be going on? If it is that, how? And what do I do about it?

There is a package called libcurl installed, and it is indeed 7.47.0, but there are no other versions of it in my repos to force. Mintupdate also only has one version, so nothing weird could have happened there. There's also python3-pycurl but that's 7.43.0 (also no other versions.) So, any ideas?

edit: Here's my dpkg.log from today https://pastebin.com/raw/NF9NYmXu
I cant see it being something other than something I installed because, otherwise, I haven't done anything today besides gently caress around in irc and check my email.


Edit: Remembered Mint comes with Timeshift now, went back a couple days and all is well, for now. Still, who knows what happened and if it will happen again :/

BrainDance fucked around with this message at 10:59 on Dec 24, 2017

BrainDance
May 8, 2007

Disco all night long!

Does anyone know if there is any way to get motion controls for a switch joycon or an 8bitdo SN30pro bluetooth controller working in Linux? Or how I can even test to see if it's working? Xubuntu 19.10.

The joycons pair, work as controllers. No motion controls as far as I can tell, I googled around a bit and everything I found led me to believe that it's just not a thing that is done by anyone yet. So, I moved onto the 8bitdo sn30pro, which advertises motion controls. It pairs just fine and I've used it plenty of times as a normal controller. Then I tried to setup motion controls in Dolphin. This page https://dolphin-emu.org/blog/2019/11/07/dolphin-progress-report-october-2019/ led me to believe that the motion control axis would just be exposed to Dolphin if they exist and then, tada, setup motion controls. But, the controller configuration just does not see anything, joycon or sn30pro.

I googled around for documentation on the sn30pro's supposed motion controls but found a giant heap of nothing. So, short of buying another controller specifically to emulate one game (at that point I might be better off buying a third Wii). I'm just not sure what to try next.

BrainDance
May 8, 2007

Disco all night long!

Hoping somebody can help me here, I'm at a loss on how to get the proprietary nvidia driver working on this computer (laptop, should be an optimus card, but gently caress that noise I just wanna use the nvidia gpu.)

It works in Ubuntu for what it's worth, but haven't had any luck with Debian sid

quote:

$ uname -a
Linux --- 5.4.0-3-amd64 #1 SMP Debian 5.4.13-1 (2020-01-19) x86_64 GNU/Linux

quote:

$ nvidia-smi
Fri Feb 7 13:45:48 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.44 Driver Version: 440.44 CUDA Version: N/A |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce MX130 On | 00000000:01:00.0 Off | N/A |
| N/A 50C P8 N/A / N/A | 0MiB / 2004MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

quote:

$ lspci | grep 3D
01:00.0 3D controller: NVIDIA Corporation GM108M [GeForce MX130] (rev a2)
output of lshw -c video

From what I can tell the nvidia card is not getting used at all, installing the driver on its own with the instructions here doesn't break anything, but the driver isn't being used, and the nvidia card isn't being used, just the lovely onboard intel one.

quote:

$ nvidia-settings

ERROR: Unable to load info from any available system
dmesg when the driver is installed but it's still in a working state
Xorg.0.log from when it's still in a working state

So, then I follow the instructions here
My xorg.conf ends up looking like this

quote:

Section "Module"
Load "modesetting"
EndSection

Section "Device"
Identifier "nvidia"
Driver "nvidia"
BusID "01:00.0"
Option "AllowEmptyInitialConfiguration"
EndSection

But, when I restart SDDM, x dies, I reboot and same deal no x.
dmesg from after it fails
Xorg.0.log from after it fails

Comment out everything in xorg.conf and it goes back to "working". I tried to let nvidia-xconfig make its own xorg.conf too but that also fails (even more spectacularly, I just get a black screen on boot but I can get out of it and get to a terminal at least.) And I tried some stuff from the arch wiki (although it doesn't seem too different) here and it makes no difference. I could keep googling, but to be honest I have no idea what is or why it's broken so it's just going to be me trying random things that probably wont get anywhere.

Anyone have any ideas?

BrainDance
May 8, 2007

Disco all night long!

6205.630 FPS on the same hardware with a distro I'm sure was using nvidia hardware (and I guess it would have to be, I don't think __GL_SYNC_TO_VBLANK=0 works on anything but the proprietary nvidia driver on nvidia hardware.)

62.191 FPS on Debian :( I'm guessing because __GL_SYNC_TO_VBLANK=0 didn't actually do anything there.

BrainDance
May 8, 2007

Disco all night long!

nvidia-smi says no running processes found while running glxgears :/

BrainDance
May 8, 2007

Disco all night long!

I "fixed" it, in the sketchiest, dumbest way possible.

Made a backup, said "gently caress it", copied 10-nvidia.conf and 11-nvidia-prime.conf from usr/share/X11/xorg.conf.d in the working ubuntu install to the same place in the broken debian install, rebooted and it works. I guess Ubuntu knows how to better write an xorg conf than I do.


If anyone cares or may want to look at it if they run into it in the future, here's what was in those conf files.
https://pastebin.com/raw/nEJgcuVn

My money is on "Option "PrimaryGPU" "Yes"" being what actually fixed it, but I'm not gonna gently caress with it anymore.

BrainDance fucked around with this message at 11:16 on Feb 8, 2020

BrainDance
May 8, 2007

Disco all night long!

The Debian Administrator's Handbook is free and an actual book/ebook, and it takes on the pretense of "You're a company switching to Debian heres how you do stuff". It does assume you're intermediate and tosses all the basics in a quick appendix. PopOS is some flavor of Debian I think, so it'd be at least somewhat relevant (but not exactly)

https://debian-handbook.info/browse/stable/

It's not as dry as sitting there just reading man pages.

BrainDance fucked around with this message at 04:16 on Feb 10, 2020

BrainDance
May 8, 2007

Disco all night long!

Right, exactly, so just sitting there reading them one after the other to try and learn everything would be..... dry.

BrainDance
May 8, 2007

Disco all night long!

You are reading way, way too much into a "don't just read the man pages" line...

Edit: I mean, youre not wrong, it's just kind of a weird reply to something that was agreeing with you.

BrainDance fucked around with this message at 16:06 on Feb 10, 2020

BrainDance
May 8, 2007

Disco all night long!

I've been really happy with Debian Sid. Yeah Sid means you lose out on the benefits of Debian's stability but really it's fine, it's been more stable than anything Arch derived I've ever tried (though that might be because I'm an idiot. And I did have an SDDM update eat a conf it shouldn't that broke X for a second, but it was an easy fix. And that mess getting the Nvidia proprietary driver working in the beginning. )

The thing that makes me side with Debian over Mint or Ubuntu is, I don't know why or how but I have had much more luck just sticking to the default repos with Debian. You'd think everything in a Debian repo would be in a Ubuntu repo too, but it's just not. All too often a solution to something in Ubuntu ends up being people recommending you install x package from repo y that I have absolutely no reason to trust.

BrainDance
May 8, 2007

Disco all night long!

Has anyone ever tried Mint Debian? I'm comfortable where I am but I'm just curious about how that all works out. Last I had heard it didn't seem as mature as the Ubuntu based Mint, but I haven't been anywhere near Mint in a long time.

I don't even remember what made me want to get away from Mint, something they did bothered but it was so long ago.

Ubuntu, and sort of by proxy Mint and probably Pop_OS, really is king of 'it just works even with weird laptop hardware'. Until a bios firmware update Ubuntu was the only thing I could get to boot without disabling secure boot on one laptop (which was impossible before the firmware update), the laptop is a Chinese model so the internet wasn't much help. I still keep that Ubuntu partition around for now because 'how does Ubuntu do it?' has been useful sometimes. But I feel like Canonicals always been the problem, there's the threat of them making some weird crappy decision that's going to go against what the majority of their users want and then I'm gonna have to deal with that if I ever want to update anything.

My reasons for thinking that are probably ancient history at this point, but can you really use Linux without holding decades of grudges over minor design choices in a distro?

BrainDance
May 8, 2007

Disco all night long!

Hey, I'm an idiot, what's wrong with my bash script? Excessive echos there because I was trying to figure out what's up

quote:

audio=`pacmd list-cards | grep 'active profile'`
analog="active profile: <output:analog-stereo>"
echo $audio
echo $analog
if [[ "$audio" == "$analog" ]]; then
pactl set-card-profile 0 output:hdmi-stereo
else
pactl set-card-profile 0 output:analog-stereo
fi

Even if $audio and $analog appear to be the same, it doesn't work and runs pactl set-card-profile 0 output:analog-stereo

BrainDance
May 8, 2007

Disco all night long!

tjones posted:

Put double quotes around your echo statement variables and you'll see why its not matching.

ex:
echo "$audio"
echo "$analog"

code:
	active profile: <output:analog-stereo>
active profile: <output:analog-stereo>
CASE 2

Yeah, this is what I get when I did that, so now $analog is " active profile: <output:analog-stereo>" instead and it works lol. Thanks everybody.

Edit: It was more clear in the post window, but yeah it's a character before active profile which I didn't otherwise see.

BrainDance fucked around with this message at 07:56 on Mar 11, 2020

BrainDance
May 8, 2007

Disco all night long!

Thanks, that's actually really helpful.

I figured there had to be a better way to accomplish what I was trying to do. Didn't know how else to go about it, though.

BrainDance
May 8, 2007

Disco all night long!

Maybe we need a shell developed by machine learning with something like Google searches by old people who don't understand how to use a search engine as the sample data.

Lowest common denominator syntax.

BrainDance
May 8, 2007

Disco all night long!

If anyones running whatever the latest version of Ubuntu is could someone give me the output of
code:
for f in /var/lib/apt/lists/*Packages; do 
  printf '%5d %s\n' $(grep '^Package: ' "$f" | wc -l) ${f##*/}
done | sort -n
It should output the total number of packages in your repos, nothing sketchy. I'm having trouble finding numbers elsewhere.

BrainDance
May 8, 2007

Disco all night long!

Thank you! I'm sure there was an easier way to find this out, but I couldn't find it lol.

BrainDance
May 8, 2007

Disco all night long!

the libdns-export1110 package in Debian Sid has had an outstanding bug for about a month now, I've just been holding it and going on with my life.

code:
critical bugs of libdns-export1110 (&#8594; 1:9.11.18+dfsg-1) <Outstanding>
 b1 - #954736 - Upgrade to 9.16.1-1 causes dhcpd to die with SIGABRT
Summary:
 libdns-export1110(1 bug)
https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=954736

From my limited understanding, it looks pretty serious. I don't want to install it and have DNS suddenly dead. But if its just been sitting there for a month, and Debian really wants me to install it (this is a new package, not an upgrade of an older one), maybe I should just go with it. Could someone take a look at it and tell me what the practical effects of dhcpd dying would be?

BrainDance
May 8, 2007

Disco all night long!

Is Moonlight still the best way to go about streaming games from a Windows machine with an nvidia gpu over a network to a Linux machine (that also happens to have an nvidia gpu, if it mattered)? Or are there new/different ways to go about doing the same thing?

Setting up an awesome home theater, got a projector with really low latency and playing games on it works great, but I didn't spend all this money on an expensive GPU to be playing only Switch games and SNES roms all day.

Debian 11, KDE, proprietary nvidia driver, the computer hooked up to the projector is very mid-range.

BrainDance
May 8, 2007

Disco all night long!

Is DKMS broken or something?

BrainDance
May 8, 2007

Disco all night long!

I do get the value, in certain cases sure whatever (though I don't understand why appimages aren't enough. I'm sure there's a reason though, I just don't know it.)

I hate the idea of having to use any of them for any particular package though. For my case, and I imagine for a lot of people who aren't using Linux in some corporate environment, it just adds complexity instead of reduces it. Now I gotta deal with 2+ different kinds of packages, and snaps/flatpacks at least perform worse than appimages/native packages, so why would I ever want that? Why do I want my browser to have to come in some other weird format? Everything works really well for me as is!

But that's just what I'd think if someone really tried to convince me to use snaps or something, because as it is now I just use Debian and don't use any snaps or flatpacks, no problem for me there.

BrainDance
May 8, 2007

Disco all night long!

Phosphine posted:

"Ubuntu, but like, without Canonical's muckups".

Which is kinda funny cuz that's how I see Mint.

Though, LMDE exists and I like that idea so much more. I'm all about Debian and I think it's a good recommendation when people specifically say "I am new and want to learn Linux. Because you might have to, you're gonna have to probably check the wiki to get your wifi working with some firmware from the nonfree repo. But it will still all probably work and be stable at least.

But if someone just wants something that works and they don't really care about learning anything, Mint just seems like the right place to be.

I dunno the goon perception of Arch, but I dont like the AUR in Arch for the same reason I don't like PPAs in Ubuntu, you put all that trust in what? Just some dude? At least with Debian there's some kind of trust. But yeah whatever you get bleeding edge packages I guess.

BrainDance
May 8, 2007

Disco all night long!

I have what's maybe a Linux problem, but it could just as likely be a hardware problem. The only way right now that I know of to narrow it down would be to either haul another computer in this room (don't wanna do that) or install Windows on the computer already in here to test it out (also don't wanna do that.)

I have this projector, a th685 that is supposed to be able to do 120hz 1080p. Connected to the projector over HDMI (and should be a good, new, shielded HDMI cable) I have an older laptop, the GPU in the laptop is an Nvidia mx130. On the laptop is Debian 11, with the proprietary Nvidia driver, and kde in xorg. The laptop itself has a 60hz screen.

xrandr shows that 1080p 120hz and 119.88 are valid refresh rates for this projector, but when I try to switch to those refresh rates for it the projector freaks out, gives me some out of range errors and then will eventually come to and "work" at 120hz, for about 5 seconds with the screen glitching out a bit before going back to "out of range."

I am pretty sure the problem isn't the projector itself, because it's been the same on 2 projectors (I had to send the 1st one back for a stuck pixel.)

I'm hoping it's just that the mx130 is too garbage to manage it, because as soon as I get a new gpu for another computer I'll be using my old parts hoard to build a small computer in this room with a 2070s to replace the laptop, but is it maybe a Linux issue? I know xorg isn't always nice about multiple displays with different refresh rates, and Debian is a bitch sometimes with the proprietary driver. Is there some way to figure out what the problem even really is that I'm not thinking of?

BrainDance
May 8, 2007

Disco all night long!

Got a weird one, not exactly sure what to do here and a quick google gave me mostly Windows advice that doesn't seem applicable here.

I got a home theater thing that I run Debian Sid on (was running stable, but I like Sid I like when poo poo breaks and I gotta learn to fix it, and there was a package that was actually too old in current stable and made it so I couldn't do some stuff, a bluetooth thing, long story but doesn't matter because it is fine in Sid.) It's a laptop with an nvidia gpu, proprietary driver (I doubt that matters here, but it might), then over hdmi to my receiver and out to a projector/5.1 speaker setup. 5.1 is setup and working with pulse, at least for the most part. In the Jellyfin client, works fine I get 5.1, in Firefox it also works fine.

But, I was using Chrome for all that stuff, I used to have trouble with Firefox and playing videos on it back in the day and just ended up using Chrome for media pc stuff. I only use Chrome for this and Firefox on every other computer I use, so I don't actually know much about what to do when Chrome breaks.

It used to be working, but I had to go with just stereo for a couple months while I replaced a receiver. Now, new receiver so back to 5.1, and no matter where I'm playing a video from in Chrome, netflix, youtube, wherever, it will only play in stereo. I feel like it has to be a Chrome problem if Firefox and everything else plays in 5.1 and this is the only thing not working. I'd blame Sid but, Chrome isn't coming from Debian's repos. So, I dunno, anyone heard of this before or have any idea what I would do to even figure out why it's not working?

It doesn't matter much since I can just use Firefox but if I leave it unfixed it's gonna bug me.

BrainDance
May 8, 2007

Disco all night long!

VostokProgram posted:

Can we calculate when Debian will run out of Toy Story characters?

There's no fixed schedule for Debian releases but I counted 109 characters on https://en.wikipedia.org/wiki/List_of_Toy_Story_characters

Cant wait for Debian Bonnie's Mom and Debian A toy robot with suction cups that allow it to stick to the wall, which plays music for the bath toys

Going by the amount of time it took for the last 10 named releases (not counting bookworm or anything named after) and taking that as an average, we have 174.8 years of named Debian releases left. Unless they make another toy story in that time.

BrainDance fucked around with this message at 07:39 on Dec 16, 2022

BrainDance
May 8, 2007

Disco all night long!

I'm trying to install cuda 11.7 on WSL2 Ubuntu
nvcc -V tells me I got cuda 11.5
code:
(base) ckg@Ryzen9:/usr/local/cuda$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0
So, I tried sudo apt --purge remove cuda, then I installed cuda 11.7
code:
wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
And uh.... yeah cuda 11.5 is still there, nvcc -V tells me the same thing, how do I get rid of cuda 11.5 and install 11.7?

BrainDance
May 8, 2007

Disco all night long!

BrainDance posted:

I'm trying to install cuda 11.7 on WSL2 Ubuntu
nvcc -V tells me I got cuda 11.5

So, I tried sudo apt --purge remove cuda, then I installed cuda 11.7
code:
wget https://developer.download.nvidia.com/compute/cuda/11.7.1/local_installers/cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
sudo dpkg -i cuda-repo-wsl-ubuntu-11-7-local_11.7.1-1_amd64.deb
And uh.... yeah cuda 11.5 is still there, nvcc -V tells me the same thing, how do I get rid of cuda 11.5 and install 11.7?

Just for the sake of always writing the answers to my problems on the very off chance someone eventually has the same problem (or more likely, I have the same problem again and forget how I solved it the first time) There were two issues here.

First is, I purged the cuda packages, but forgot that there's nvidia-cuda-toolkit which was very much what I needed to remove.

Second thing is, I had to set the path to cuda 11.7 for it to work and for nvcc -V to say that's what I'm using (and for everything else to use it)
code:
export PATH=/usr/local/cuda-11.7/bin:$PATH
export CUDA_HOME="/usr/local/cuda-11.7"
export LD_LIBRARY_PATH="/usr/local/cuda-11.7/lib64:$LD_LIBRARY_PATH"
export PATH="/usr/local/cuda-11.7/bin:$PATH"
Got my where I needed to be

BrainDance
May 8, 2007

Disco all night long!

This is barely a Linux question so sorry if this is way too far off, but it's a thing done in Linux and it's software that only works in Linux (DeepSpeed doesn't really build on Windows), the solution I need is some kind of Linux software (or any advice!) and I have exhausted every other option I can think of to ask for advice. And, in short, I know nothing about csv files, other than they're not complicated things and I don't know what's possibly breaking but I don't know how to check. So, can anyone tell me how to check and figure out what's different between these two files? I assume there's some tool somewhere in Linux that can get me all the data in the world about a csv file.

So, the whole issue, I am attempting to fine-tune GPT-Neo. And this is where I don't exactly know where to start when describing this process because it's a long, complicated thing. But the complicated stuff works, pretty much everything works, I can train it and there's no problem, at least with the sample data included in the repo (it's here https://github.com/Xirider/finetune-gpt2xl)

So, you get 2 text files, one for the training data one for a validation file. Then, using a script in the repo you convert them to csv files, then you train them. They are interpreted by datasets when you actually do the fine tuning. If I take the example Shakespeare text files, use the script to convert them to csv files, then finetune it all works fine. When I take my own text files though and convert them to csvs datasets has a seizure and it fails with "IndexError: index out of bounds"

code:
Traceback (most recent call last):
  File "/home/ckg/github/finetune-gpt2xl/run_clm.py", line 478, in <module>
    main()
  File "/home/ckg/github/finetune-gpt2xl/run_clm.py", line 398, in main
    lm_datasets = tokenized_datasets.map(
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/dataset_dict.py", line 471, in map
    {
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/dataset_dict.py", line 472, in <dictcomp>
    k: dataset.map(
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 1619, in map
    return self._map_single(
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 186, in wrapper
    out: Union["Dataset", "DatasetDict"] = func(self, *args, **kwargs)
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/fingerprint.py", line 397, in wrapper
    out = func(self, *args, **kwargs)
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/arrow_dataset.py", line 1977, in _map_single
    writer.write_batch(batch)
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/arrow_writer.py", line 383, in write_batch
    pa_table = pa.Table.from_pydict(typed_sequence_examples)
  File "pyarrow/table.pxi", line 1559, in pyarrow.lib.Table.from_pydict
  File "pyarrow/array.pxi", line 331, in pyarrow.lib.asarray
  File "pyarrow/array.pxi", line 222, in pyarrow.lib.array
  File "pyarrow/array.pxi", line 110, in pyarrow.lib._handle_arrow_array_protocol
  File "/home/ckg/anaconda3/envs/p39/lib/python3.9/site-packages/datasets/arrow_writer.py", line 100, in __arrow_array__
    if trying_type and out[0].as_py() != self.data[0]:
  File "pyarrow/array.pxi", line 1067, in pyarrow.lib.Array.__getitem__
  File "pyarrow/array.pxi", line 549, in pyarrow.lib._normalize_index
IndexError: index out of bounds
[2023-02-16 18:50:28,897] [INFO] [launch.py:318:sigkill_handler] Killing subprocess 424
[2023-02-16 18:50:28,897] [ERROR] [launch.py:324:sigkill_handler] ['/home/ckg/anaconda3/envs/p39/bin/python', '-u', 'run_clm.py', '--local_rank=0', '--deepspeed', 'ds_config_gptneo.json', '--model_name_or_path', 'EleutherAI/gpt-neo-1.3B', '--train_file', 'train.csv', '--validation_file', 'validation.csv', '--do_train', '--do_eval', '--fp16', '--overwrite_cache', '--evaluation_strategy=steps', '--output_dir', 'finetuned', '--num_train_epochs', '1', '--eval_steps', '15', '--gradient_accumulation_steps', '2', '--per_device_train_batch_size', '4', '--use_fast_tokenizer', 'False', '--learning_rate', '5e-06', '--warmup_steps', '10'] exits with return code = 1
I've tried to give it two different sets of my own training data and it fails in the same way for both. They are very different, one is short with short lines arranged in verses, the other is long with only long single lines (chat logs.)

But I'm pretty sure there is something with the way I'm formatting these text files that is making it not make the csv file in the right way. So, I'd think if I could go and see what's different between the two besides the actual text I could maybe figure out what I'm doing wrong, but how do I even do that? All I've done is "cat train.csv" and go "mhm, yep, that looks like text"

Heres the sample text: https://cz0.au/Feku.txt
and the sample csv: https://cz0.au/6G-8.csv
That works

Here is an example of my training text: https://cz0.au/_HXe.txt
And my training csv: https://cz0.au/FnQ4.csv
That don't work.

Google has gotten me nowhere besides some jerk 2 years ago saying he had the same problem then just "nvm it was a problem with my text, fixed it" without saying what the problem was.

BrainDance
May 8, 2007

Disco all night long!

JLaw posted:

Shot in the dark but your input does contain a lot of unusual character codes, i.e.

That actually seems like it might be something. The other text, it's ripped from chat logs. The chat logs are from a Chinese chat program though that sometimes formats things weird (like weird apostrophes.) I've built a whole process for cleaning them up, like, for the chat logs
awk '/username/ {f=NR} f && NR==f+1' file > file.txt to extract my messages
grep -vE "^[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}:[0-9]{2}" file.txt > output.txt to try to filter out any loose screennames (they have timestamps, sometimes the previous awk would grab someone mentioning me and then copy the next line which would be a username)
grep -P -v '[\x{4e00}-\x{9fa5}]' file.txt > output.txt to remove Chinese characters (the Chinese chat program, it saves stickers posted as a little Chinese text flag thing)
sed 's/http[^ ]*//g' file.txt > output.txt to strip it of urls
grep -v '^[[:space:]]*$' file.txt > output.txt to empty lines
sed -i 's/^[ ]*//g' file.txt to get rid of some leading white space
awk '{print "responseuser1: " $0;}' file.txt > output.txt to add the trigger word for the AI
sed -e 's/$/<|endoftext|>/' -i file.txt to add the terminator (though, this is gonna need to be reworked)

It's a whole event and I wouldn't be surprised if that process broke something, or if the data just had those weird character codes. Is there anything I could do with sed or something that would just strip it of any weird character codes? I'm not sure how to word that, but you know, stuff that isn't just... normal?

BrainDance
May 8, 2007

Disco all night long!

Saukkis posted:

I once had to use find to search any files with unusual characters, I suspect grep or sed would support similar character set searches.

code:
find . -maxdepth 1 -regextype posix-egrep -not -regex '[[:alnum:][:punct:][:space:]]*'

Thanks, this got me started on the right path, managed to do it with sed
code:
sed 's/[^[:print:][:space:][:punct:]]//g' file.txt > output.txt
And then it still didn't work, lol. But by that point I was like "ok, the text is clean, this makes no sense"
So, then I just took the working Shakespeare example and appended the daodejing to the bottom of it and tried that, and it started training.
Then I deleted all the Shakespeare stuff and left the daodejing... and it trained. Which left only the validation file.

I was using one example for each for the validation file. So one verse of the daodejing or one message from the chat logs. The Shakespeare validation file is longer because Shakespeare just is longer than a verse from the very short daodejing. And, in the end, basically it turns out there is a minimum length for the validation file. This is not documented anywhere, because the documentation for this is all kinda garbage, but yeah that seems to be what it was.

Adbot
ADBOT LOVES YOU

BrainDance
May 8, 2007

Disco all night long!

jaegerx posted:

No there are a number of 3rd party apps you can download to do this just as easily.

There are and I hate that. Every single problem in Windows is "a number of 3rd party apps" which are of varying quality and often come packed with a bunch of other junk you don't need.

Though the Windows implementations of the core utilities are mostly fine. There are still some things about Windows that makes me usually go for a Linux install to do that stuff instead of using them.

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply