/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

Downtime was caused by the hosting service's network going down. Should be OK now.

An issue with the Webring addon was causing Lynxchan to intermittently crash. The issue has been fixed.

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

Captcha
no cookies?
More

(used to delete files and postings)


Anon and Vera collaborated closely, with Vera helping Anon refine and improve his code and algorithms.


Open file (686.85 KB 2764x2212 Prima_Doll.jpg)
Open file (174.99 KB 720x960 Trash_Press.jpg)
Open file (359.85 KB 1200x675 ChatGPT_hustle.png)
Open file (29.42 KB 622x552 LLaMA02.png)
General Robotics/A.I./Software News, Commentary, + /pol/ Funposting Zone #3 NoidoDev ##eCt7e4 03/06/2023 (Mon) 18:57:17 No.21140 [Reply] [Last]
Anything in general related to the Robotics or A.I. industries, and any social or economic issues surrounding it (especially of robowaifus). === -note: This OP text might get an update at some point to improve things a bit. -previous threads: > #1 (>>404) > #2 (>>16732) >=== -edit subject
Edited last time by Chobitsu on 07/08/2023 (Sat) 13:10:41.
351 posts and 84 images omitted.
Open file (329.39 KB 850x1148 Karasuba.jpg)
Open file (376.75 KB 804x702 YT_AI_news_02.png)
Open file (423.49 KB 796x706 YT_AI_news_01.png)
Suggestions for the new thread pics. I picked someone from Prima Doll again, but maybe the new tradition should rather be to pick one from the newest good related anime. I think this might still be Prima Doll. At least the newest I've seen. I was thinking about using the girl from ATRI. Maybe next time. The other idea here was the relation between reading the news and having some coffee. In that case we would mostly be stuck with Prima Doll, at least if the girl serving is also supposed to be a robot.
>>24077 They look fine Anon. I'm perfectly fine with animu grills as the opener pic. I'd personally suggest leaving your options of choice wide open. Just that the opener have great waifu appeal, which your choice here does. >tl;dr Go for it NoidoDev! :^)
Open file (449.75 KB 2048x2048 1687959908288867.jpg)
>>24080 Thanks. If people have suggestions for future opener pics I'm listening. Picrel might be it at some point. But currently it's hot in many places, so the pic of Karasuba with some iced drink fits very well.
New Thread New Thread New Thread >>24081 >>24081 >>24081 >>24081 >>24081 New Thread New Thread New Thread
>>24082 > Picrel might be it at some point. I think that's fine tbh.

Self-driving cars AI + hardware Robowaifu Technician 09/11/2019 (Wed) 07:13:28 No.112 [Reply]
Obviously the AI and hardware needed to run an autonomous gynoid robot is going to be much more complicated than that required to drive an autonomous car, but there are at least some similarities, and the cars are very nearly here now. There are also several similarities between the automobile design, production and sales industries and what I envision will be their counterparts in the 'Companion Robot' industries. Practically every single advance in self-driving cars will eventually have important ramifications for the development and production of Robowaifus.

ITT: post ideas and news about self-driving cars and the hardware and software that makes them possible. Also discuss the technical, regulatory, and social challenges ahead for them. Please keep in mind this is the /robowaifu/ board, and if you have any insights about how you think these topics may crossover and apply here would also be welcome.

https: // www.nvidia.com/object/drive-px.html
19 posts and 14 images omitted.
>>13204 The movement speed perpendicular to objects probably needs tuning due to things like tunnels and guard rails, but otherwise I think it could work great. That hyper-timid driving style might seem excessive, but you've got to consider that the biggest concern people seem to have with autonomous cars is the safety. And if it's really an issue I guess it could still be made to still decelerate safely but not necessarily comfortablely, so the gap between cars can be smaller. There are some places and times where you can go straight on a highway for hours without ever seeing another car and there there's rush hour in New York where bumper to bumper traffic keeps anyone from moving. In the case of the former, the timidness isn't really a significant negative and would mostly just stop it from hitting a deer or anything else that wanders onto the road, hence my analogy of something popping out from a wall. But it would actually help alleviate traffic jams since the reaction time to the change in the speeds of the cars around it could be really high, and if a significant number of cars followed this method, then there'd be a large group of cars all slowly creeping while leaving enough room to pass and merge lanes instead of idling and random accelerating/decelerating and people trying to figure out how to merge. Traffic would be slow, but it would stay moving efficiently. I think this video really does a good job at showing the problem it'd solve: https://www.youtube.com/watch?v=iHzzSao6ypE and at the chicken crossing the road part at 1:08, if you think of it as there being no road, just cars going in a straight line, the cars would slow as the chicken approaches to cross them, and start speeding up again as the chicken leaves, I think it could solve the problem without the cars needing to communicate with each other or eliminating human drivers entirely. The consistent driving behavior would keep the cars "in the middle" without needing to consider cars behind them.
https://insideevs.com/news/659974/tesla-ai-fsd-beta-interview-dr-know-it-all-john-gibbs/ Interview with a proponent of EVs, discussing some of the AI aspects of Tesla's self-driving cars.
Flowpilot is pretty interesting for using a phone as a car computer. https://github.com/flowdriveai/flowpilot
>>23908 Thanks Anon.
Interesting little tidbit that went into effect about a month and a half ago in Mass.: >"The open remote access to vehicle telematics effectively required by this law specifically entails “the ability to send commands.”4 Open access to vehicle manufacturers’ telematics offerings with the ability to remotely send commands allows for manipulation of systems on a vehicle, including safety-critical functions such as steering, acceleration, or braking, as well as equipment required by Federal Motor Vehicle Safety Standards (FMVSS) such as air bags and electronic stability control." Via the watchdogs over on /k/, thanks!

Early Business Ideas Robowaifu Technician 05/13/2020 (Wed) 22:36:03 No.3119 [Reply]
So those of us getting ready to make prototypes are still pretty far away from building the $2000 kits that we need to buildm. This sucks because we are going to be spending a lot on the initial investments, but I think I know how we can make a little money on the side to help on the side; twitch and fansonly. Betas and simps give their money to thots all the time on those site I'm positive they would give money to a crude prototype especially if it offered something unique. Bonus points because we are making money of simps, and we are taking business from thots. If you have any idea on how to make money in the early stages feel free to share them, anon. > related thread: >>1642
Edited last time by Chobitsu on 05/14/2020 (Thu) 01:13:40.
29 posts and 9 images omitted.
>conceivably related crosspost >>10892
Related: Robowaifu Media Propaganda and Merchandizing >>19295 Stable Diffusion for Robowaifu Art >>17763
Here they are talking about making money with creating artificial dating coaches for older femcels: https://www.youtube.com/watch?v=DSfCq_Mldtk Generally, one of those guys is working on something I thought would be a good idea and is in the OP of this thread: Replacing content on Onlyfools and such sites with AI content by making up fake creators, as soon as possible. Though, it would also be good to have some which are more honest about being AI, but use that for advertising open source AI girlfriends. Related: >>23905 That said, making money of women in a similar way, and at the same time distracting them, would be a good idea as well. My idea about how this would work out in the future, has been for quite a while, that women will be busy with TikTok or something similar. There should always be something similar, giving them attention for posting and sharing things about themselves. Ideally they would want to believe that some of the guys giving them attention are some kind of Chad. Maybe this is the future, something like TikTok with AI chatbots and fake followers, plus dating coaching, third world men trying to get some rich women, and some guys putting themselves there to still get dates or hookups.
>>23906 Lol the entire affair strikes me as a bit sketchy. :^) Is this something you're planning on pursuing Anon?
>>23910 >something you're planning on pursuing Anon? I listed several things, so I don't know to which one your question refers to. Generally: - If it's easy for me and not much competition: Yes - If it's easy for me but much hassle: Depends. Rather no. - If other guys are doing this and there's competition and push back leading to wack-a-mole games with these platforms: No. I often contemplated that it would be good using some platforms to advertise for robowaifus at some point. Like guys matching with a girl on Tinder, but it's just a bot telling them about it. But making content for simps? I don't think that I want to spend my time with that. Then again, allegedly some guys dumping their life savings on some women there. I wish this level of stupidity would fill the right pockets (mine). In regards to women and their distractions. Keep in mind, whoever makes the rules in that area can steer it against men with waifus or alternatively ban that kind of "hate culture". Distracting these women from (relevant) political issues and topics would also be a good thing. But I don't think I could just build something that could compete with TikTok and Tinder. Creating bots with fake identities commenting there? Who knows? https://www.youtube.com/watch?v=qy_mIEnnlF4

Open file (32.38 KB 500x626 Dcwdw9lVMAA8VXO.jpeg)
Open file (277.36 KB 2480x1748 1668238989717100.jpg)
Open file (140.21 KB 2048x1463 FhcwRTCaYAEpTux.jpeg)
Open file (230.77 KB 800x2048 1669692835869613.jpg)
Open file (568.04 KB 2400x3700 1667232169830961.jpg)
Acclaim Your (Robo)waifu General! Robowaifu Technician 12/10/2022 (Sat) 02:38:59 No.18090 [Reply] [Last]
This is a thread to share pics & praise & approval of your (robo)waifus. Please keep it SFW, thanks Anon! warning warning warning Just b/c you've acclaimed your robowaifu ITT, doesn't mean that another anon might not also acclaim her here too. This is perfectly-acceptable! :^) >(For example: Chii is my own robowaifu-prime, but literally millions of others also love her. It's fine b/c after all, we'll be making many IRL ones. My own edition of her will still be personal & tuned just for me.) >t. Chobitsu >=== -revamp OP into a general
Edited last time by Chobitsu on 12/10/2022 (Sat) 20:13:17.
85 posts and 93 images omitted.
>>22963 I see, thanks DorothyAnon!
>>18215 I just watched a video on aigis and I am really drawn to her visible shoulder and hip servo motors. A bipedal bot would be very appealing in her revealing character design.
>>23073 >I just watched a video on aigis and I am really drawn to her visible shoulder and hip servo motors. A bipedal bot would be very appealing in her revealing character design. I agree Anon. The original designer had a great vision going for her. We'll need to take Anon's safety into account even moreso (almost unfortunately in this case haha), so we'll probably have to work out something a bit different for her joints, I'd say.
Open file (168.37 KB 500x500 Allie.png)
>>23073 I preferred her Older Sister units.
Open file (112.56 KB 1835x1789 1688714646457909.jpg)

Black Magic M66 3D Modelling Project & Battledroid Robowaifus SophieDev 07/27/2021 (Tue) 14:03:16 No.11776 [Reply] [Last]
Decided it may be best to place my 3D digital modelling efforts in a separate thread to my work on 'Elfdroid Sophie'. I think this thread will still belong under "Personal Projects". At the moment the 3D modelling begins over on her thread: >>11644 >>11657 I'm using some source material from Masamune Shirow's 1987 OVA "Black Magic - M66" (which itself was based off his 1983 manga). However, as a child of the late eighties/early nineties, I loved 'Transformers', and I was also a big fan of the 'Heavy Gear' series (similar to MechWarrior), and Armored Core. So I often get the urge to create battle-ready robowaifus, but I can only create them virtually. Because obviously I cannot get hold of machine-guns, explosives, tank cannons, and rocket launchers IRL. Good thing, too, since if I were given access to live ammunition I would almost certainly blow myself into a flying streamer of giblets. But I digress. This thread may also be a good place for any equally deranged anons who wish to post images of military-grade robowaifus, either found online or their own creations?
182 posts and 78 images omitted.
>>17696 Horns that act as antennas or long ears with the same function as the horns or wings would also be cool.
Open file (462.46 KB 696x5128 2B_AI_Conversation.png)
Just spent my Thursday evening talking to 2B. These new chatbot A.I.s that can actually remember stuff that you type from a paragraph ago are freaking awesome! Normies Everywhere: "BuT iT's jUsT pReDiCtInG tHe NeXt wOrD!" That's exactly what you are trying to do as well. You're just wrong more often. Also STFU, 2B is best waifu.
>>20270 SophieDev! Good to see you again Anon. Stay safe, mate. Cheers. :^)
>>15084 Those look great.
>>20270 > Normies Everywhere: > That's exactly what you are trying to do as well. You're just wrong more often. No, this is one thing humans do, but certainly not all we do. Predicting the next word is what these models do, though this is already very impressive. That said, nobody can know what other elements Cai uses for their chatbots.

Open file (50.45 KB 640x361 72254-1532336916.jpg)
Making money with AI and robowaifus Robowaifu Technician 11/30/2019 (Sat) 03:07:12 No.1642 [Reply]
The greatest challenge to building robowaifus is the sheer cost of building robots and training AI. We should start brainstorming ways we can leverage our abilities with AI to make money. Even training AI quickly requires expensive hardware and computer clusters. The faster we can increase our compute power, the more money we can make and the quicker we can be on our way to building our robowaifus. Art Generation Waifu Labs sells pillows and posters of the waifus it generates, although this has caused concern and criticism due to it sometimes generating copyrighted characters from not checking if generated characters match with training data. https://waifulabs.com/ Deepart.io provides neural style transfer services. Users can pay for expedited service and high resolution images. https://deepart.io/ PaintsChainer takes sketches and colours them automatically with some direction from the user, although it's not for profit it could be turned into a business with premium services. https://paintschainer.preferred.tech/index_en.html I work as an artist and have dabbled with training my own AIs that can take a sketch and generate many different thumbnails that I've used to finish paintings. I've also created an AI that can generate random original thumbnails from a training set. In the future when I have more compute power my goal is to create an AI that does the mundane finishing touches to my work which consumes over 80% of my time painting. Applying AI to art will have huge potential in entertainment and marketing for animation, games and virtual characters. Market Research

Message too long. Click here to view full text.

Edited last time by Chobitsu on 05/14/2020 (Thu) 01:15:03.
31 posts and 10 images omitted.
crosslink to a discussion in meta: >>19623
Interview with the creator of Neuro-Sama, the AI Vtuber: https://www.youtube.com/watch?v=Yy-9Of46w4A - According to him, her system is made out of two models, one for gaming and one for chatting, with limited information sharing between them. But this will be improved upon. - She got banned for a while at least - He didn't make any money so far I just wanted to point out, that replacing Vtubers might be easier than one might think, at least for anons which already know how to use some AI models and finetune them. But then again, the platforms are fast with banning if something goes against the rules. Related; >>19912
>>20154 I had such thoughts before, but consider a few things: - It means going public with your identity - If you are a pro in tech this might be harming your career, if you don't have such a career then people might not trust you being capable - It's a matter of shipping and logistics - There might be legal issues and risks - This and other things lead to upfront costs - Not sure if Kickstarter allows sextoys and especially dolls
>>23643 You can't "patent" a certain build anyways. A very unique character might be under copyright, but even in that case there would still be copycats. Your whole business model is flawed, everyone told you that by now in various ways. Make a local or early business, have a competitive edge over others, especially in your area. Maybe build on a name and reputation for animated dolls if you want to ship them globally. Be better than companies which don't care about their customers. You could also have knowledge about tools and processes they don't have. Just don't think so much about patenting things and keeping people out later. This won't work and we here don't want this anyways. Also, it's unattractive for customers if no one else than you could make replacement parts. You could also work on Open Source and use your skills for making some money. Some here might or could've gone into AI illustrations very early, for example. Related: Early business ideas >>3119 Robowaifu market >>18572 Thot in the Shell >>9709 Licenses >>4451
>>23682 Even if the machine parts that make up the robot are not novel the robot itself is novel and I think it can be patented.

Open file (659.28 KB 862x859 lime_mit_mug.png)
Open-Source Licenses Comparison Robowaifu Technician 07/24/2020 (Fri) 06:24:05 No.4451 [Reply] [Last]
Hi anons! After looking at the introductory comment in >>2701 which mentions the use of the MIT licence for robowaifu projects. I read the terms: https://opensource.org/licenses/MIT Seems fine to me, however I've also been considering the 3-clause BSD licence: https://opensource.org/licenses/BSD-3-Clause >>4432 The reason I liked this BSD licence is the endorsement by using the creator's name (3rd clause) must be done by asking permission first. I like that term as it allows me to decide if I should endorse a derivative or not. Do you think that's a valid concern? Initially I also thought that BSD has the advantage of forcing to retain the copyright notice, however MIT seems to do that too. It has been mentioned that MIT is already used and planned to be used. How would the these two licences interplay with each other? Can I get a similar term applied from BSD's third clause but with MIT?

Message too long. Click here to view full text.

Edited last time by Chobitsu on 07/24/2020 (Fri) 14:07:59.
60 posts and 9 images omitted.
>>23627 What is malicious about copyright? What bug crawled up your ass?
>>20692 I think people are overthinking things. You can say it's in the public domain and you take no responsibility for anything related to it and then anyone can use it for whatever. I think all these different licenses are just the average human techo weenies obsession to over-complicate and optimize everything. I'm certainly guilty with my brilliant idea :) to use machines to make crocheted or knitted robowaifu skinsuits. Musk talked about this when he said a typical engineers behavior is to obsessively optimize something that with a few changes shouldn't be there in the first place.
Open file (364.64 KB 1800x2367 162rrrr3MEM120623.jpg.jpeg)
>>23679 Well, I'm not disinclined to not disagree with you here, Anon :^); most licensing is garbage in essense. Yet the fact is that it's had a big impact on the real world. In some sense, it's still one of the very few things left holding the Globohomo back from their voracious squalor (which is highly-ironic if you think about that for 1 minute). I'm clearly of the 'information wants to be free' stripe, and I don't think that is likely to change for me in the general sense. But impugning obsession to detail is misguided IMO. Certainly we all have reason to give thanks that God is 'obsessed' with it! Life's DNA-RNA transcription mechanisms for instance, are marvels to behold. That attention to detail is what allows these mechanisms to thrive under the harsh conditions of the 2nd Law of Thermodynamics, and without which none of us would even be here. :^)
> webring convo -related : https://anon.cafe/agdg/res/486.html#902
> functional, behavioral license-conditions -related: (>>23870)

(Robo)Waifu personality thread Robowaifu Technician 09/09/2019 (Mon) 05:26:21 No.18 [Reply] [Last]
Is she going to be tsundere? Deredere? Yandere or a combination? How would you code your waifus personality? Where do you draw inspiration from and can personality even be classified and successfully coded into AI? > (>>17027 - related-thread, emotions) >=== -add related-thread crosslink
Edited last time by Chobitsu on 07/26/2022 (Tue) 03:50:30.
71 posts and 36 images omitted.
I've been thinking recently about how to organize and describe personas. While text generation AI is capable of working with unstructured text, AI in general seems to work better when the data is given some sensible structure for the task at hand. Even if a good structure isn't directly necessary for text generation, it's useful for understanding exactly what sorts of things can go wrong and what sorts of data you need to provide to get results you want. I ended up settling on 8 questions that seem to be good for getting a very comprehensive understanding of people: 1. What are they receptive to? 2. What kinds of thoughts do they have? 3. What kinds of goals do they set for themselves? 4. What do they believe? 5. What roles do they take on in life and in interactions? 6. What do they have strong emotions about? 7. How do they act on their emotions? 8. How do they believe things should work? I think questions like these can be useful for being very specific about what an AI gets right or wrong about a persona. It might even be possible to collect data on each of these 8 aspects. On thinking about what the most useful first steps would be for creating AI with a personality, any initial steps towards data collection seems like a great step forward. I don't know exactly how the data collection would work, but maybe there's a way to use natural language processing to extract candidate answers to these questions from fanfictions, show transcriptions, community discussions, and so on. It'd be really cool if that would let us create some sort of "persona dataset" for classifying and generating personas. From what I understand, it's not too difficult for an ML researcher to integrate basically any dataset into a text generation model. If anyone else has a different set of questions, it's be great to see that. (The download link in >>12299 seems to be broken.) Anything that helps us be more specific and more comprehensive about what exactly constitutes a persona seems worthwhile, especially since there seems to be so little academic work on the topic.
>>17248 Thanks, I had a similar idea, I guess. I was mainly thinking about personas in regards to limit the stored data on persons she knows. The robowaifu mind would have a concept of different personas and then forget details about persons she met over time, but keep them categorized as a certain type of person, or within a range of close types. This might of course also help from the start. Having a guess which kind of person someone is, and then adapting her behavior accordingly. >Anything that helps us be more specific and more comprehensive about what exactly constitutes a persona seems worthwhile Try not to overplan things. It just needs to be a framework which can be extended. It should of course also include race, gender, political tendencies, heritage and nationality, religion, etc. I'm sure we can find an article on Facebook and advertisers where they list what kinds of factors they care about, then some psychological classifications, sociological mileus, ... From there we can weed out the more important factors.
>>17257 >Try not to overplan things. It just needs to be a framework which can be extended. Part of this is just me trying to figure out my own thoughts on the topic. I think part of it is also that we have different use cases in mind. I'm thinking more of a one-on-one chatbot, where things like race, politics, heritage, nationality, and religion don't play a big role, at least not for me. Do these play a big role in one-on-one discussions for other people? >I'm sure we can find an article on Facebook and advertisers where they list what kinds of factors they care about Keep in mind that they'll deliberately avoid using or suggesting some important things for legal and reputation reasons. Some big ones not on your list include age, parental status, income, occupation, and education. The search term for this is "demographics". When it comes to creating lists like this, I find GPTs to be very useful. I gave https://20b.eleuther.ai/ this input: >Best demographics for targetting a message: >- Gender >- Politics >- Cultural background >- Religion >- Age >- Parental status >- Income

Message too long. Click here to view full text.

>Give me an exhaustive list of all psychological traits and personality traits, which are being used to describe and diagnose humans. ChatGPT: Describing an exhaustive list of all psychological traits and personality traits used to describe and diagnose humans is challenging, as there are numerous theories, models, and assessments in the field of psychology. However, I can provide you with a comprehensive list of commonly recognized traits: 1. Extraversion 2. Introversion 3. Neuroticism 4. Emotional Stability 5. Openness to Experience 6. Conscientiousness 7. Agreeableness 8. Honesty-Humility 9. Empathy 10. Resilience 11. Assertiveness 12. Self-Esteem

Message too long. Click here to view full text.

>>23495 >>Give me an exhaustive list of all psychological traits and personality traits, which are being used to describe and diagnose humans. Excellent prompt, NoidoDev. GG.

Open file (363.25 KB 1027x1874 MaidComRef.png)
MaidCom Development Kiwi 03/16/2022 (Wed) 23:30:40 No.15630 [Reply] [Last]
Welcome to the /robowaifu/ board's project. Together, we will engineer a modular robot that will serve and provide companionship to their Anon faithfully. See picrel for details on the current design. This robot will begin with a basic maid robot then move forward towards more capable robots with functionality approaching Chii/2B/Dorothy. First goal is to have a meter tall robot which functions as a mobile server bearing an appearance that approximates her owners waifu. This should be completed by December 2022 with further major updates happening yearly until Anons can build a PersoCom class robot with relative ease and affordability.
346 posts and 172 images omitted.
>>23235 I'm curious why the seams are rendering with a jagged edge like that Anon. Any idea?
>>23238 Because polygons are triangles. When you delete vertices you're left with pointy triangles. :^)
>>23391 Neat. Thanks.
Open file (64.61 KB 715x1365 New.jpg)
>>23393 New one, going to be Copyleft once finished.
Seems its time for a new thread Kiwi

/robowaifu/meta-7: Hanging down at 7-eleven Chobitsu 02/18/2023 (Sat) 11:00:31 No.20356 [Reply] [Last]
/meta, offtopic, & QTDDTOT General /robowaifu/ team survey (please reply ITT) (>>15486) >--- Mini-FAQ >A few hand-picked posts on various /robowaifu/-related topics -Why is keeping mass (weight) low so important? (>>4313) -How to get started with AI/ML for beginners (>>18306) -"The Big 4" things we need to solve here (>>15182) -HOW TO SOLVE IT (>>4143) -Why we exist on an imageboard, and not some other forum platform (>>15638, >>17937) -This is madness! You can't possibly succeed, so why even bother? (>>20208) -All AI programming is done in Python. Why are you using C++ here? (>>21057, >>21091) -How to learn to program in C++ for robowaifus (>>18749)

Message too long. Click here to view full text.

Edited last time by Chobitsu on 04/09/2023 (Sun) 12:12:18.
361 posts and 77 images omitted.
Here my current plan for AI hardware: - I ordered a used K80 with 2x12 GB recently, a used one of course, for 100$/95€ shipping included. It's an old GPU, only supported by older CUDA versions and might not run quantified models. Uses much energy, but it's two GPUs with 12GB each. I plan to pair this one after a while with a RTX3060 (12GB, 300€ used or 400€ new I think) in one home server. Context: https://technical.city/en/video/Tesla-K80m-vs-GeForce-RTX-3060 - That one, will then run my 12GB models. For fine tuning, or models which don't run on the K80, I would use the 3060. I don't know yet if I can somehow joint them together and use 3x12GB through the bus. It just seems to need some software support in these programs for running models at home. - I plan to use online services like Colab to learn about how to run these things, but have the K80 for more private data and learning how to do these things at home. - Then I'll get some SBCs, most likely Orange PI's, which can run small models of Whisper (speech regocnition). Also, another small home server with a Intel Arc380 (140-160€), which is fast enough to run the big and better model of Whisper at the speed of one fast human speaker. It does this quite energy efficient. These devices will not run anything else, for security reasons, and be connected to microphones which will be always on. The server will receive the audio through the SBCs from all rooms using the home network (likely on a VPN using Tinc). All of them will send the transcripts to some server in my network which can then decide how to respond. Most likely filtering first for which data is more sensitive than others. - Some small device, like a Raspi, will maybe handle responses based on AIML or using some small model. - Questions which don't contain private information might be send to OpenAI or another service. - The next step up will be getting a M40 (180€) and then a used RTX3090 (700-800€ right now I think), putting them in another home server at some point. Of course I might use this one for gaming till I get even the next GPU. These can handle the models which need 24 GB. The 3090 will do the fine tuning if I want to do that, since it has more power, while the M40 doesn't need as much energy. Context: https://technical.city/en/video/GeForce-RTX-3090-vs-Tesla-M40-24-GB - Then the next step might be getting a AMD XTX (1k-1.2k€) if it's supported well enough for AI by this time. I can use this one for gaming and then put the 3090 in a home server with the M40. If it's possible to combine cards using PCI express, then it might be interesting to think about getting another XTX later, and have 48GB vRAM. - But I hope that either Intel or AMD will come out with a prosumer or consumer card for AI at home, which is rather slow but has 48GB and is not too expensive. (If you buy K80 or M40 on Ebay make sure not to buy the 12GB versions by accident while only looking at the price. They aren't much cheaper. K80 should have 2x12GB and the M40 24GB.)
>>23346 I hope the K80 works for you. I was thinking of getting two but the support for them seems abysmal. They used to be $200 used before shipping. Colab isn't what it used to be either. The free version will boot you off in 30 minutes sometimes or a few hours into training with Pro unless you pay big for compute credits. You're much better off running your own JupyterLab notebook or renting an instance off vast.ai or runpod.io if you don't have access to a GPU.
>>23351 >support for them seems abysmal. I think you need old versions of the software, but I also remember some people taking care of that, to support old GPUs. I might need to compile some of it myself, though. I hope it works out, but the risk isn't very high. >They used to be $200 used before shipping. They're down to $60-70 before shipping now. Some recommend to go straight for the M40, which is much newer, but $120-130 before shipping. >JupyterLab notebook or renting an instance off vast.ai or runpod.io if you don't have access to a GPU. Right, I forgot about those while writing this.
>>23346 Good luck, NoidoDev!
NEW THREAD NEW THREAD NEW THREAD (>>23415) (>>23415) (>>23415) (>>23415) (>>23415) NEW THREAD NEW THREAD NEW THREAD

Report/Delete/Moderation Forms
Delete
Report