|
Does anyone have experience with running TensorFlow on a Pi? I'm trying to set up a text generating API on my Pi 4. It works fine when I run it from my desktop PC but when I try to run it on the Pi I get an error that the 'tensor.contrib' module is not found. I know tensor.contrib was removed in TF 2.0 but I made sure to install 1.14 from pip, which is the same version I'm using on my desktop. I've found instructions on how to compile TensorFlow yourself from the source on a Pi but I'm not sure if that would actually solve the problem, although I do see people saying that the official TF packages for the Pi are for some reason lovely.
|
#
¿
Dec 28, 2019 17:58
|
|
|
|
| # ¿ Apr 24, 2024 14:03 |
|
|
Has anyone tried installing the 64 bit version of Ubuntu Server on a Pi 4 yet? I see back in July there was a bug where it was limited to 1 GB of RAM but I'm not sure if that's been fixed.
|
|
#
¿
Jan 5, 2020 04:40
|
|
|
So I did end up installing Ubuntu 19.10 on it Saturday night but I've been having a recurring problem with it since I did. About once a day, I'll be unable to SSH into and being unable to access the database/local network service running from it. When I try to SSH in using myname@ubuntu I get an error that the Pi's ECDSA key has changed and because I have strict checking enabled will not be able to connect. Removing the Pi from my known_hosts file stops this error, but then I begin getting a Permission Denied (publickey) error. If I instead try to use myname@the.pi's.ip.address I get an error that the resource is temporarily unavailable. Hard rebooting the Pi fixes this (I have not tried plugging it in to a monitor in this state to see whether it is actually on before I switch the power off and back on). I don't believe the power is flickering and causing it to shut down, as all the digital clocks in my apartment would also be reset. Can anyone think of an explanation for this other than the Pi powering off?
|
|
#
¿
Jan 8, 2020 04:02
|
|
|
accipter posted:Could you have two devices fighting for the same IP address? I did see that as a possibility when I was searching for an answer but at least at the time that I checked after rebooting none of the devices on my network had the IP that the Pi normally does. I did go ahead and reserve it for the Pi though just in case.
|
|
#
¿
Jan 8, 2020 05:36
|
|
|
Does anyone have experience using a Pi 4 for object recognition and can you speak on its accuracy/speed? One of our semester-long projects is literally shark tank but for a product related to pharmacy and I was considering trying to use my Pi to make a prototype pill counting camera. The project is more about giving us experience in pitching/planning a business idea and we're not expected to actually make our thing, but I think it'd be cool if I could. Actual pill counters can run for $5k-$15k at 99.9% accuracy and I wanted to see how feasible it could be to make something that would cost much, much less. From what I've read the Pi 4 should be capable of object recognition at about 8 FPS, or it could possibly even go up to the refresh rate of the camera if a $75 USB tensor processor is used. My other uncertainty is how the training of the model would work, as pills/capsules can take on many different shapes and sizes.
|
|
#
¿
Jan 27, 2020 21:50
|
|
|
Does anyone have recommendations for a good case for the Camera Module V2 (just the camera, not the Pi as well)? I ordered the camera and case off of CanaKit, but the case is apparently not compatible with the V2 module cameras despite not mentioning this at all and being the only camera case they sell. The hole for the lens in the case is just off enough that the lens rubs against it instead of fitting into it, and so the module as a whole sticks out too much to actually close the casing.
|
|
#
¿
Feb 5, 2020 00:02
|
|
|
Has there been any progress on 64-bit OS stability on the Pi 4s? I tried Ubuntu Server when it first became available but was having some issues with it such as it randomly dropping all networking ability (it would still be running and I could plug a monitor/keyboard in, but SSHing or accessing it via the IP/port I had a service running on didn't work). Would like to try again mostly so I can use the newer versions of MongoDB.
|
|
#
¿
May 27, 2020 19:26
|
|
|
Are you using the 64 bit Raspberry OS beta by chance? I know VLC is broken on it but I don't know to what extent. Either way, apt upgrade shouldn't brick your system though.
|
|
#
¿
Oct 27, 2020 20:06
|
|
|
I started trying to gently caress around with TensorFlow on my Pi 4 again but now that I'm using the 64 bit OS beta I'm getting an ELFLCASS32 error in Python. It seems that pip only wants to install the 32 bit version of Tensorflow on a Pi. I don't see where anyone has already built Tensorflow from its source for this Pi/OS, so I just started the process myself.quote:After 33 hours compilation, hopefully, you will see with the following screen. Wish me luck 
|
|
#
¿
Oct 27, 2020 23:23
|
|
|
Pilchenstein posted:I have no idea, I'm just using whatever the default option was - how do I check that? It's only available to download from a post on their official forums, so you would have had to seek it out for that. ante posted:Hah, are you actually trying to compile on the Pi? Cross compiling on your desktop is usually just a couple extra compiler flags Yup! I've literally never done something like this before and found a guide that walked me through doing it on my Pi. It's not being used for anything else right now so other than the wait time (and the possibility that it could always just not work anyways after it's done) it's no loss to me to just let it run.
|
|
#
¿
Oct 28, 2020 04:22
|
|
|
King Bahamut posted:I just set up a standard pi 4 raspbian desktop from canakit, got my VPN up and running, downloaded some SNES ROMs because I’m a 40 year old stereotype, and ran into a wall. Files are in a 7zip format (.7z) but xarchiver doesn’t recognize them, through the GUI or command line. Man on xarchiver says it goes by the file header rather than the extension—do I need to search for a specific archive of ROMs created on a Linux file system? Can I convert or am I missing something dumb I don't know if it's in the Raspbian repos but the Linux 7zip package is called "p7zip".
|
|
#
¿
Oct 29, 2020 20:39
|
|
|
 
|
|
#
¿
Oct 29, 2020 21:32
|
|
|
Well I got the model running and it took my Pi 4 about 58 minutes of sitting at 99% RAM usage to generate the following:quote:2015 Last time I tried to mess with this I was looking into converting the GPT-2 model to a TensorFlow Lite model but many of the operations it uses weren't supported by Lite.
|
|
#
¿
Oct 30, 2020 01:27
|
|
|
namlosh posted:This is actually cooler than suggested by the content lol Nope, still GPT-2. I haven't looked at the file sizes/complexities of GPT-3 but I'd imagine it would take even longer to generate. I trained the model on some movie scripts and YouTube video subtitles on my desktop PC, but I had to use the 355M base (the second smallest) because anything bigger uses too much VRAM for my 1080 Ti to handle. I was originally trying to do this as a replacement for a Markov chain in a Discord bot but the generation time doesn't really lend itself to that. I might move down to the 124M model and see if that's any better. e: Actually, the GPT-3 model isn't even available for download. It's only through an API and you have to sign up for a waitlist to get access. And I'm guessing "shitposting chat bot" isn't a valid reason to put on the application. e2: also it's $100 a month minimum after the first 3 months lmao i vomit kittens fucked around with this message at 00:43 on Oct 31, 2020 |
|
#
¿
Oct 30, 2020 21:14
|
|
|
CarForumPoster posted:If you need access to GPT-3 I might be able to find the inside line. I know a few people who have access pretty much to gently caress around with it. I got to use a very early beta GPT-3 based chat bot a few months ago and it was pretty amazing. Definitely would blow any Markov based method out of the water. I don’t recall what the look back was on it but it was a lot. I guess it doesn't hurt to ask because yeah I'd love to mess with it provided I don't run the risk of somehow getting someone else's access revoked or cost them anything. Just to be up front though, it's 100% not for any sort of productive or thought-provoking application, it'd literally just be relegated to trying to get it to say stupid poo poo every once in a while in a Discord of ~75 people.
|
|
#
¿
Nov 1, 2020 12:28
|
|
|
I wouldn't buy one myself but it's definitely a cool idea and would probably be great for a kid getting into programming/electronics. It also has a faster processor than the regular Pi 4s, 1.8 GHz vs. 1.5. I don't know how significant of a jump that is but I was considering buying a second Pi 4 so it'd be nice to see a faster CPU show up on them.
|
|
#
¿
Nov 2, 2020 11:13
|
|
|
Nohearum posted:Has anyone ever built a docker image based on Raspbian? I'm running LibreELEC on my pi and in order to run moonlight (streaming games from PC) I have to run it inside a docker container. I found a Raspbian image on dockerhub but I'm kind of sketched out since its not an official image and could be mining bitcoin behind the scenes for all I know. There are official Raspbian images but the last time I looked into it the most recent one was Jessie/8 while the current version was Buster/10. I took the Jessie one and updated it to Buster myself.
|
|
#
¿
Jan 1, 2021 06:54
|
|
|
|
| # ¿ Apr 24, 2024 14:03 |
|
|
Does anyone else using PiHole have issues with it locking the filesystem after losing power suddenly? I have two Pis, a Zero and a 4. I've experienced the same problem on both, but only when they're running PiHole; the one that's doing other things works fine after a power outage. The only way to fix it seems to be reimaging.
|
|
#
¿
Jan 20, 2021 19:27
|
|