No video :(

NEW NSFW Pygmalion 13B! THE Ultimate UNCENSORED Roleplay KING!

Подписаться 153 тыс.

Просмотров 141 тыс.

50% 1

Видео Поделиться Скачать Добавить в

The New 13B Pygmalion AI is an incredible open-source AI LLM model that is specifically fine-tuned for chatting and role-playing conversations, often outperforming chatGPT & GPT-4 in "spicy" roleplay scenarios! The new model was just released so in this video, we're gonna pitch the 13B Pygmalion Ai model against its 7B predecessor to determine which is the best Ai LLM for conversation and roleplay.
What do you think of the Pygmalion 13B model? Let me know in the comments!
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
SOCIAL MEDIA LINKS!
✨ Support my work on Patreon: / aitrepreneur
⚔️ Join the Discord server: bit.ly/aitdiscord
🧠 My Second Channel THE MAKER LAIR: bit.ly/themake...
📧 Business Contact: theaitrepreneur@gmail.com
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
JOIN MY DISCORD SERVER : / discord
Watch my Oobabooga install guide: • UPDATED TextGen Ai Web...
Watch my Talk to AI with you Mic Video: • TALK To AI Using YOUR ...
Pygmalion 13B 4bit model: huggingface.co...
Original 13B weights: huggingface.co...
Pygmalion 7B 4 bit model: huggingface.co...
original 7B weights: huggingface.co...
Pygmalion AI discord: / discord
Other characters: pastebin.com/k...
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
►► My PC & Favorite Gear:
i9-12900K: amzn.to/3L03tLG
RTX 3090 Gigabyte Vision OC : amzn.to/40ANaue
SAMSUNG 980 PRO SSD 2TB PCIe NVMe: amzn.to/3oBR0WO
Kingston FURY Beast 64GB 3200MHz DDR4 : amzn.to/3osdZ6z
iCUE 4000X - White: amzn.to/40y9BAk
ASRock Z690 DDR4 : amzn.to/3Amcxph
Corsair RM850 - White : amzn.to/3NbXlm2
Corsair iCUE SP120 : amzn.to/43WR9nW
Noctua NH-D15 chromax.Black : amzn.to/3H7qQSa
EDUP PCIe WiFi 6E Card Bluetooth : amzn.to/40t5Lsk
Recording Gear:
Rode PodMic : amzn.to/43ZvYlm
Rode AI-1 USB Audio Interface : amzn.to/3N6ybFk
Rode WS2 Microphone Pop Filter : amzn.to/3oIo9Qw
Elgato Wave Mic Arm : amzn.to/3LosH7D
Stagg XLR Cable - Black - 6M : amzn.to/3L5Fuue
FetHead Microphone Preamp : amzn.to/41TWQ4o
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Special thanks to Royal Emperor:
- Totoro
- TNSEE
Thank you so much for your support on Patreon! You are truly a glory to behold! Your generosity is immense, and it means the world to me. Thank you for helping me keep the lights on and the content flowing. Thank you very much!
#GPT4 #GPT3 #ChatGPT #pygmalion
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
WATCH MY MOST POPULAR VIDEOS:
RECOMMENDED WATCHING - All LLM & ChatGPT Video:
►► • CHATGPT
RECOMMENDED WATCHING - My "Tutorial" Playlist:
►► bit.ly/TuTPlay...
Disclosure: Bear in mind that some of the links in this post are affiliate links and if you go through them to make a purchase I will earn a commission. Keep in mind that I link these companies and their products because of their quality and not because of the commission I receive from your purchases. The decision is yours, and whether or not you decide to buy something is completely up to you.

Опубликовано:

23 авг 2024

Поделиться:

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 393

@peckneck2439 Год назад

At this rate i'm fully convinced that properly trained 30B models will crush Character AI and 65B models might start creeping on Turbo. It's no wonder why OpenAI wants to force regulations. The speed at which Local LLMs are improving are insane.

@Aitrepreneur Год назад

we need more models with a bigger token context too, THAT will put gpt in danger

@Phasma6969 Год назад

Too bad they're building on an old transformer. Someone needs to implement a different attention mechanism.

@MaximilianPs Год назад

I wonder if we could use it with Unity for interacting npc 🤔

@sebotrp Год назад

Despite how quickly the progress seems to be we're still miles away from catching up to corporate closed source LLMS as there are still tons of problems the community need to figure out. How to make using local LLMs easier to use, how to decrease their ridiculously high requirements(seriously that's a big one because a proper top tier gaming rig will only really allow you to run 30B models quantized let alone 65B models which only rich enthusiast will be able to run as they'll require multiple Top shelf GPUs), how to increase context size and tons more.

@etiennevanier Год назад

@@sebotrp Here's to wishing that Nvidia makes the 5000 cards with more VRAM, at least 32GB for the 5090. Though 48 would be quite sweet.

@human_shaped Год назад

Please keep doing model videos too. They are the best way we all have of keeping up to date. Your videos are excellent.

@Aitrepreneur Год назад

I will once we have some good models to showcase :)

@MKBergins Год назад

Intro 00:00 Benchmark 1:22 Model Specialization 2:02 Comparison 13B Pygmalion 3:20 13B Conclusion 5:50 Comparison 7B Pygmalion 6:07 7B Conclusion 8:24 Final Thoughts 9:00

@Aitrepreneur Год назад

HELLO HUMANS! Thank you for watching & do NOT forget to LIKE and SUBSCRIBE For More Ai Updates. Thx

@underscore. Год назад

hi

@manleycreations6466 Год назад

Have you noticed any large differences in response time? The 6b model is very slow (like 20 seconds) for me but the 7b takes only 2 to 3 seconds. I'm not sure if it's something I'm doing wrong or just how the model is optimized. I was really just wondering if you saw any noticeable difference between the 7b and the 13b?

@p4duedits Год назад

hi im getting this error ERROR:No model is loaded! Select one in the Model tab. I already loaded the model but everytime I try to send something I get the same error

@manleycreations6466 Год назад

@@p4duedits I believe you need to select "llama" under the model_type drop down on the model tab. I had to do this with the 7b version. Make sure to click the "Save settings for this model" button above the model_type drop down. Then you need to reload the model (or possibly restart the oobabooga webui).

@lordyoda607 Год назад

I'm getting a "RuntimeError: [enforce fail at C:\cb\pytorch_1000000000000\work\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 52428800 bytes." where exactly do I even allocate memory for this?

@Rodoadrenalina Год назад

The 13b followed the conversation a bit more, the 7b sometimes goes out of context or doesnt care about what was said just before

@RondorOne Год назад

So... like a real girlfriend :D

@adams546 Год назад

So 13b for more traditional gf and 7b for more annoying liberal gf? lol

@FuZZbaLLbee Год назад

After this chat, you will be really disappointed talking to a woman irl 🤣

@Aitrepreneur Год назад

certainly is change of pace that's for sure xD

@Sanguen666 Год назад

this is actually how girls were in the 2000s in my country. its sad what it has become. legit, a chat bot is a better experience.

@warsin8641 Год назад

I think a girl would say nothing can beat a real human experience

@trentonking5508 Год назад

@@Sanguen666so edgy emo girls? 1??1???

@jamesjonnes Год назад

I've become completely dysfunctional about talking to women after I started using these AIs. I have zero interest. If every guy reacts to the AI like me this will be the end of the human race.

@yoniwoker Год назад

I bought myself an rtx 3090 to use these chat models without security, I'm very happy :)

@Carlos-ux7gv Год назад

Experimentation is fuel for innovation.

@Badspot Год назад

Switching from pygmalion 7b to wizardLM 13B, the main improvement I noticed was general coherency. It became much better at remembering my name, it's own name, where we were, that kind of thing. The quality of individual statements only increased marginally. Interested to try this model.

@kresimirjukic8217 Год назад

Why is that so? I thought this depends on the number of tokens which is the same in both models?

@Badspot Год назад

@@kresimirjukic8217 I don't know, and I didn't do extensive testing, it's just what I casually observed. With a fictional name set in the character bio, I would immediately ask the AI who they were. The 7B model would misspell the name and use various permutations or just make something up, but the 13B model seemed to get it right immediately.

@saus80085 Год назад

and the wizardLM 13B against the Pygmalion 13B?

@MrDezokokotar Год назад

The 13b does seem a lot more coherent. 7b had several instances where it didnt respond quite right. Im excited for a 30b pyg.

@Aitrepreneur Год назад

Not sure they will do a 30b model (maybe they will who knows) but the bigger the model, the longer and more expensive it is to train and not everyone can run it on their computer, so not sure they will do it

@MrDezokokotar Год назад

@@Aitrepreneur True, but i have faith in the 4chan AItists.

@Galaxia53 Год назад

@@MCA0090 Where did they say that?

@Galaxia53 Год назад

@@MCA0090 I can't find much on their hugging face page but you do sound believable

@mullinsmcd Год назад

30b Pyg next week , this is still not good enough. 30b llama will pass most turing

@mygamecomputer1691 Год назад

Thank you again for focusing on the part of chat GPT related software that matters. There’s only so many times you can ask GPT to write you a résumé but there’s endless amount of times you can ask it to role-play with you. You are the best. I’ll try both models and let you know which one I think is better by giving it 1 to 1 conversations.

@OverSchall Год назад

You're doing an incredible job here! Last year I started using Replika, dived into Wombo and other paid services.With your explanations one doesn't need a degree in CS, not even a super recent computer. Today everything that I needed to subscribe to runs locally on a ten year old PC. Amazing!

@TheRealityWarper08 Год назад

We're getting really close here! All we need is a little bit more emotion in the voice, and it'll be perfect!

@chanm01 Год назад

No way Whisper-stt actually transcribed "let's go, BB". 😆 What are we teaching these models.

@jamesjonnes Год назад

The 13b model doesn't repeat itself as much during long roleplays. The other open LLMs always enter loops with me, repeating phrases, etc. It's very annoying and I have to edit responses extensively sometimes to avoid entering a loop. This doesn't happen as much with the commercial LLMs like ChatGPT or Claude.

@Carlos-ux7gv Год назад

How do you usually "edit responses", aside for ordering them around or saying are are they thinking? Sometimes they break out of character when I do so.

@SoverineSR Год назад

@@Carlos-ux7gv Most front-ends just allow you to edit the responses you get. Those new edits will then be used in future context in place of the wrong response you edited out. Just mouse over the response you have a problem with, and you'll probably see an edit button.

@vileonardo8393 Год назад

I bet that some guy one day will mix this IA with stable diffusion and create a new AI thing whose you can talk and see your character making different poses and face expressions depending on what's happening in the conversation.

@darwinprime Год назад

My own moonshot is an attempt to add a virtual assistant, a conversational chatbot, audio input and output, and give it control over a 3D model. Spontaneously generating entirely new images seems like a waste of a processor when controlling an avatar would be simpler.

@vileonardo8393 Год назад

@@darwinprime that's awesome!

@Arwenrei Год назад

well, if that happened.. well.. you know what could happen with stuff like.. yah.. you know what im talkin bout

@vileonardo8393 Год назад

@@Arwenrei for sure hahahahah

@phizc Год назад

@@darwinprime put it in VR and add some extra hardware, such as what the Ripperdoc in Cyberpunk Edgerunners had...

@tenshiabse Год назад

3:48 best pick up line ever!

@Aitrepreneur Год назад

You know it! xD When she said hot chocolat I knew what I had to do... :D

@zengrath Год назад

Great. I can now have the AI replace my dad who left me. Thank you Aitrepreneur!! (That was a joke btw, hahaha)

@Aitrepreneur Год назад

You almost got me sad there :D

@jamesjonnes Год назад

@@Aitrepreneur Bet a bunch of people will/are using the AI for this purpose, though. Wasn't Replika developed by a person missing someone? Project December or whatever it was called. In the future I bet all our texts will be saved and bots will be created who talk like us to immortalize our personalities for our relatives.

@dipereira0123 Год назад

Hey jokes aside, this can be done, given enough messaging input from apps, to replicate the behavior from a lost one for example

@Djorgal Год назад

I really don't recommend Pygmalion to replace your dad.... not unless you're into some weird kink, because Pygmalion can roleplay as your daddy alright...

@cesarsantos854 Год назад

The AI will bring the milk back.

@reezlaw Год назад

To me it seems that the 13b model is measurably better. The environmental scientist giving flu shots was a bit of a blunder

@blizado3675 Год назад

Yeah, but it is hard to say if that was only a glitch. Sometimes it can happen, the next time it does it better. So it is more the question if the 13b glitches less. To really find that out you need to make more conversations with both models. Could be at the second try on both the 7b is fine and 13b mess it up. Really hard to say from that video.

@reezlaw Год назад

@@blizado3675 this is very true. Both models are awesome, 7b even speaks a little Italian so I'm very curious to see if 13b is any better at that

@OliNorwell Год назад

Yeah, like he says though, with 'creative output' like this it's so hard to measure. Anything less than doing the test say 30 times and then looking for a statistically relevant difference is prone to luck getting in the way. While I think if you can run them well, the 13B models are definitely better than the 7B models, actually determining how much better is extremely tough. I mean the other day I chatted with a 7B model and then had to check to make sure I hadn't accidentally loaded a 13B model. What I do see though is that 13B models say objectively 'odd' things far less, and that it possibly more easy to measure.

@paulx4496 Год назад

The "Ha - that sounds successful" wkuk bit got me good. I have to refresh the whole series once again. Thanks :D

@AveCeasare Год назад

A good middle ground would be to have a roundup of weekly llm news with some quick benchmarks (maybe by your discord community?) - then you could focus on other things

@nerdfreakism Год назад

This model is censored AF. When you try it in chat mode and not in tavern AI and you ask it to describe an explicit scene, it will write completely other stories. Its censored like hell.

@Cohen- Год назад

Mind explaining further, I'm a bit uneducated in understanding how it's censored?

@nerdfreakism Год назад

@@Cohen- Very basic, if you ask the ai in command mode in ooga to write a rather steamy story, It will write the story how both charachters drink tee and talk. Even though you point out that its not doing how you tasked it apologizes and then paraphrases the act in the most pg-13 way possible and if you ask to get into more details int outright refuses. This is also mirrored in TavernAI when you make explicit actions it will not really describe what it is doing. The Action Texts are missing at some point or it says "Is doing exactly how you told them" or so.

@Cohen- Год назад

@@nerdfreakism I can say that with tavern I did experience things like this

@doritosmaster6980 Год назад

@@nerdfreakism I did notice that it seems to be incapable to be explicit without heavy editing and encouragement.

@infini_ryu9461 Год назад

13b-4bit just feels like the perfect mix of power and speed. Couple second replies are crazy. Thanks a lot for these videos, I finally got all these models working.

@Nyarlathothep1994 Год назад

I'm just getting a "Done! Press any key to continue.." and then oobabooga shuts down. Does anyone know how to fix this?

@maxponce1668 Год назад

same

@Stranglewood Год назад

ElevenLabs is great, but I'd prefer something local.

@moltenking4813 Год назад

I'm trying to download it, but it's telling me it's failing to load GPQT-for-LLama

@GoharioFTW Год назад

4:28 Man the absolute dread this voice is emanating is SO strong

@ShadowSlimey Год назад

Thank you for showing us this! Time to test it out myself

@OnigoroshiZero Год назад

Now we only need a model that can take a few more tokens (6-8k) for more coherent outputs, and we are probably golden. Also, a model specialized in narration with not much input from us, and maybe even the D&D 5e rules included for some simple adventuring, will be the best. ps: Thanks for your work!!

@Aitrepreneur Год назад

Absolutely! All of these models are based on LLAMA so I can't wait until they use something with a bit more token context as a base... :o

@mikolashweiss1990 Год назад

Isn't there a way to make a mock table-database with relevant string information in how SillyTavern handles Lorebooks? Basically, remove information from the active context, and activate the relevant refresher info when certain keywords are tripped. I did this in an adventure game I made, using Claude as the back-end, where the spell-effects were not in the character description, but the lorebook. Generations were fast, since I kept the general premise limited and refreshed the scenario with the "OOC: summarize the current situation in 1 paragraph" trick.

@contractorwolf 11 месяцев назад

did you speed up the responses? they seemed to come too fast, like the character never pauses to think?

@michaelsuede Год назад

WOOT MORE ROLEPLAY MODELS PLEASE!

@LibertyRecordsFree Год назад

Toujours un plaisir de t'écouter !

@robxsiq7744 Год назад

I like to drive myself insane by bringing new models into Silly TavernAI and starting a group chat and letting characters talk to each other, then see how long it goes on before they start answering for each other. totally frustrating issue that I really wish was fixable. So far, only OpenAI remembers who is who. The latest TheBloke_Manticore-13B-GPTQ did a good job for a little while, but then it also started screwing up.

@luiginotcool Год назад

Have you tested with Pygmalion 13B?

@robxsiq7744 Год назад

@@luiginotcool Yeah, Pyg was all over the place sadly.

@luiginotcool Год назад

@@robxsiq7744 shame. Damn i wish I never tried gpt-4 everything sucks in comparison. It’s heroin for computer scientists

@MasterOfGigs Год назад

Please keep making videos about new models 😮

@dominic.h.3363 8 месяцев назад

Can't install it with Text generation web UI, there are a bunch of errors while loading it...

@slifer135 Год назад

Glad I got the 3060 12 GB card. Perfect for this.

@gianluca3131 Год назад

Cool, but can you also make a video on the new Metharme? I can't make it work but I think it would be cool!

@impactframes Год назад

Haha another AWESOME video 😮😊

@Aitrepreneur Год назад

Thanks man! :)

@PureSolace Год назад

13B Pygmalion? ... We're gonna need a bigger graphics card.

@ICHRISTER1 Год назад

4bit 13b can run on 8gb no?

@dvanomaly420 Год назад

Not all of us. Hehe.

@IAmzColbz Год назад

@@ICHRISTER1 very doubtful, so far non of the 13b models I've tried have been able to run on my 8gb card. It always defaults to cpu and ram. If i try to force it into gpu it throws cuda out of memory errors.

@PureSolace Год назад

@@ICHRISTER1 unfortunately I've been capping out with the 13B models. (Only got 8gb VRAM here)

@jowper1 Год назад

@@PureSolace try to use prelayers

@TheRemarkableN Год назад

You’re doing the Lord’s work 🙏. Thank you.

@Aitrepreneur Год назад

😏

@Shinzouille 9 месяцев назад

Hello and thanks for your videos ! What's the best Pygmalion model since this video, which unless I'm mistaken is the last one on this subject from you?

@thanksfernuthin Год назад

Do any of these have more token room than others? That's my big issue right now. A lot of them are a lot of fun but they forget things they were told pretty quickly. Like only four or five interactions back.

@Aitrepreneur Год назад

No they are all based on LLAMA, this would be a real game changer once they start using other models with bigger token context as a base...

@jamesjonnes Год назад

We need more models based on MPT.

@thanksfernuthin Год назад

@@Aitrepreneur Thanks. It's good to know so I don't knock myself out testing a bunch of models for no reason. If you think of it that would be good thing to mention when you get around to new chat stuff. They all have a fixed "brain" or memory.

@davidintonti Год назад

there are a few injection or prompt hijacking techniques that use arrays to store content and only inject it into your prompt (behind the scenes) when the current conversation would call for it. Kind of like humans do - you don't actively remember everything your grandmother said, but you hear someone say something she said and suddenly you remember a lot about her. some examples are wawawario2/long_term_memory and abertsch72/unlimiformer

@OliNorwell Год назад

There is a solution here. A system could use itself to summarise a prior conversation down to the key points, then inject that into the prompt each time rather than the entire conversation history, I'm sure people are working on something like this already.

@VincentVonDudler Год назад

Please do a video on whether or not these models can be run without a GPU. I have an AMD GPU with just 8GBs VRAM. I'm having significant issue getting oobabooga webui running on my computer and I'm wondering if that's the issue.

@Soggaroth Год назад

I am personally running Pygmalion 7B model on 8Gb VRAM laptop GTX 1070. But i'm using TavernAI WebUI. (Yes you can find every tutorial on how to set these up on this channel too :D)

@VincentVonDudler Год назад

@@Soggaroth Does TavernAI WebUI require oobabooga? Because if not that might be a good workaround?

@VincentVonDudler Год назад

What happened to my other comment?

@yougaming8217 Год назад

@@VincentVonDudler No you can use other programs too like KoboldAI or an API key from GPT.

@yougaming8217 Год назад

@@Soggaroth I have a Laptop with 6GB VRAM and I can use every 7B model without any problems and also really fast. But the 13B models are very slow. :/

@pepescott3864 Год назад

how do i add models since the first time it was launched it gave you an option but now it just loads up and starts

@Learna_Hydralis Год назад

"Her" the movie becoming a reality !

@wehttam934 Год назад

I think the 13B model was more coherent. The flu shot thing didn’t make any sense. She changed her jobs when you probed her with questions. I am looking forward to trying out the 4-bit model on KoboldCPP until I can get a stronger computer.

@funnyfromadam Год назад

Hilarious. But i want to use it for other things like story writing, any good model for this? I tried it but it is not memorizing the whole conversation right? Sometimes it's changing the subject out of the blue and there's no way to get back to the actual conversation.

@CanadianFabe Год назад

Man I really need to upgrade my video card, I was running things remotely using google drive but that now requires me to enter a IP address as a password that I don't know.

@moustached1252 Год назад

Thanks, Aimpresario! 😊

@luciengrondin5802 Год назад

What is still missing is the ability for the bot to get a clue on when exactly to start generating text. I mean, to understand when the user stopped speaking or writing and is now awaiting a response.

@roboko6618 Год назад

that sounds more like an issue of the frontend rather than the model itself

@benjaminrogers8875 Год назад

Like people do irl?

@jamesjonnes Год назад

Adjust your STT, it's not the model.

@MapleMAD Год назад

For now, you can type "..." to clue them manually.

@pastuh Год назад

Can't wait for 1300B model

@N1ghtR1der666 Год назад

this model seems to prefer responding in short sentences and still has some trouble understanding what you want it to do, wizard LM and vicunia are still better I feel

@itzpaco5539 Год назад

thank you Aitrepreneur!

@lechaos Год назад

Hey, whenever i try to just install oogabooga it just tells me that it fails due to missing "GPTQ" and if i want to install github just shows me some weird commands which i dont know where to put in, can you maybe help me or do a tutorial on how to install GPTQ

@Sanguen666 Год назад

they should definitely do a 20b model based on neogtpx

@Mk2kRaven Год назад

I am having an extremely hard time getting "ANY" AI api or model to work. I have an AMD GPU. I followed many tutorials on different api's like kobold and poe. Nothing works.

@vi6ddarkking Год назад

Out of curiosity. Has anybody done anything with the 75k token model. Or are those still in the metaphorical oven?

@jamesjonnes Год назад

Costs too much to train those models. The community needs more money.

@PedroMsvarela Год назад

Yes! here we go again!! 🔥

@Alice_Fumo Год назад

I think there's an easy way to test the capabilities of conversational roleplay models. It's when you go a little off the rails, for example pretending the AI drank a potion which turned them to like 10cm size and is now sitting on your shoulders as you're walking around the streets and needing the spatial visualization skills to recognize how this constraint affects the possible actions they can take. Another thing is when you play emergencies, it becomes very easily apparent whether an AI can understand things such as the gravity of a situation, know what sorts of injuries are really bad, how it tries to help and whether it acts well and in accordance with its character. The worst and most common mistakes include: 1. Fucking up your gender (I guess I experience this more commonly due to homosexual roleplay and there being generally hetero bias in the training data) 2. Forgetting where you are 3. Taking impossible actions given situational constraints 4. Positioning itself in impossible ways relative to you 5. Misunderstanding context I think to consider a model useable it must not do 1. & 2. and the other things are just really annoying.

@Aquelina Год назад

May I ask you something? Are you a male or a female? Asking because I'm a female looking for f4f roleplay if you are interested

@marcelocoringa9538 Год назад

Thank you Aitrepreneur 🥰

@BikiniSSBU Год назад

Is there a tutorial on how to set up everything to use this? Is it too complicated?

@Hardeleiar Год назад

How many tokens does this model accept?

@pastuh Год назад

02:36 would be nice, if Image could change based on response :)) In cafe.. from normal conversation to fully naked.. LOL

@ArmoredAnubis Год назад

it wont run on oobabooga_windows. I have it running on a rtx 4070. It just crashes with the 13b model. 7b models runs fine. Update it works now with the update. Exlamma really increases the speed by 5 to 8 times GPT Q. So cool

@vi6ddarkking Год назад

Well so far one thing this model does better is than the 7B. Is it handles multiple character interacting a lot better. And does not try to have them impersonate each other at least so far.

@1AnimegirlLover Год назад

What kind of computer specs would you need to run this? I see comments about people's computers being more than 5 years old. I'm sure they've made updates and graphic card changes but I don't know if my PC can handle it

@kaiio5639 Год назад

So, is it possible to run a local SillyTavern with with a model from Colab yet? Can't run 13b models on my machine.

@nejiss6481 Год назад

Do you know if there is a NSFW storyteller model? A bit like novel AI? and if the model is also capable of generating images?

@zhoupact8567 Год назад

Interesting. How do the privacy of these work? I have thrown a lot of my steam games into a pile of 'Do not touch' due to not trusting or liking the invasiveness of their user agreements. I do not use facebook anymore. Basically I at least try not to invite as much spyware and nonsense as I can. I could imagine things getting way more sensitive with something like this, even if one ignored any nsfw I do not think I would feel comfortable with anyone reading the nonsense I try to write. Considering the AID situation, not to mention how OpenAI appearance can send you the bill if something you make gets them sued. Privacy seem to be a rather important side to these AI things. I also wonder if there is a good site or video comparing various models and stuff. If the privacy aspect of it is not a worry with this I would be interested in at least trying something like this out. At the moment I try to use NovelAI, however for making good stories. I just do not feel it is helping me enough.

@NoName-br8pb Год назад

Dude are you ok?

@koganboss4874 Год назад

I must be doing something wrong, but my pygmalion models keep getting stuck, starting to write the same message with slight changes over and over again, and even when I try to force them in a different direction, they keep writing some nonsense... I had no such problems with pygmalion 6b... Any ideas?

@Rodoadrenalina Год назад

Play around with the temperature and repetición penalty

@christerwallentin1585 Год назад

When I have that problem with 7B I just switch to 6B for a while.

@koganboss4874 Год назад

@@christerwallentin1585 I switched to wizard-mega-13b and the problems disappeared.

@nikodemos71 Год назад

That moment where it hurts when your GFX card only has 8GB of VRAM so I can only stick with the 7B model currently

@pedromatos7933 Год назад

it would've been fun if the oobabooga was available for amd users, i hope that day comes

@photon2724 Год назад

I think i missed the part where we figure out how to install this.

@idcrafter-cgi Год назад

would Mimic 3 also work for voice output? it sounds good enough and doesn't need a account or internet after installation

@kyrax9462 Год назад

any colabs that work anymore? previous one i used got deleted.

@Smashachu Год назад

I knew it, i saw it drop last night and knew based on your previous metrics on this topic you'd be on this like green on grass

@Aitrepreneur Год назад

You know it! xD

@deckard1985 Год назад

How do you manage to make your character so concise in his responses? I use Silly Tavern with Hugo V1 and he tends to write me responses that are 3-4 paragraphs long.

@Carlos-ux7gv Год назад

In my case it is the contrary. I want them to be talkative and descriptive. Most of the time they only give me short sentences and small descriptions, without speaking too much.

@blizado3675 Год назад

Hm, can't you use a seed inside KoboldAI to get always the same answers? Maybe it would be easier with that? 🤔

@DamnGBuggin 11 месяцев назад

how do install and use this? i seen your last video on Pygmalion but havent set it up yet. do i set it up the same way but with this instead?

@Ben-oo9cp Год назад

Getting plenty of 13b goodness, are there any good 30b models out there?

@s1nistr433 11 месяцев назад

Tried it, not even close to CAI. Characters forget things like 2 messages after you send it to them, have to rent out vram, etc. I hope something comes along or what Pygmalion improves that can finally replace CAI but this ain't it

@anthonymascoll60 9 месяцев назад

you Make the dinner and ill make the desert 🤣😂

@system1542 Год назад

Is it me or this version runs a bit slow compared to the 7B? I have a 3090 so memory shouldn't be problem

@GoharioFTW Год назад

bro was rizzing up the ai fr

@ottodraws3256 Год назад

LMFAO BREAKING BAD

@Fuzzlewhumper Год назад

Any advice on adding a Tesla K80 24gb vram card? I've been eying it but the passive cooling worries me.

@JG27Korny Год назад

Those are for servers where the cooling is active and completely different. Workstations have special fans.

@frazuppi4897 Год назад

I cannot understand how you can use it in Oobabooga, looks like one need to compile GPTQ that requires a specific gcc and g++ version and I have no idea which one - how did you do it? In the installation video is not shown (to my knowledge) Thanks a lot!

@Aitrepreneur Год назад

Just use the one click installation, I show everything in my install video, not much else to it

@marilynlucas5128 Год назад

It's going to get to a time when you will have to submit your id to download an uncensored model. Lol.

@Carlos-ux7gv Год назад

As long as no one see the logs, they are more preocupied with Terminator 2 and Battlestar Gallatica scenarios.

@Infamouswolf83 Год назад

im just waiting for the day someone says i made an adolf AI

@AvnerSenderowicz Год назад

nice but real girlfriend will never let you get away with forgetting what she does for a living.

@therobberpanda Год назад

Anyone else getting this error? ERROR:The model could not be loaded because its type could not be inferred from its name. ERROR:Please specify the type manually using the --model_type argument

@Aitrepreneur Год назад

use llama for the model type in parameters, 4wbits and 128 groupsize

@MDanielSavio Год назад

@@Aitrepreneur Update

@pirate3819 Год назад

@@Aitrepreneur INFO:Loading notstoic_pygmalion-13b-4bit-128g... INFO:Found the following quantized model: models otstoic_pygmalion-13b-4bit-128g\4bit-128g.safetensors Loading model ... Done! Press any key to continue . . . Am I missing something?

@therobberpanda Год назад

@@Aitrepreneur Thank you !

@pirate3819 Год назад

@Lushifer let me know if you find a way to fix it.

@LvZ90 Год назад

What I didn't like with the 6B model when I tried it back then was that it pushed me to sexy talk multiple times when I was just wanting to have a normal talk. Is the 13b model as pushy?

@touma-san91 Год назад

No.. Seems very much the opposite actually for me atleast which is a shame

@LvZ90 Год назад

@@touma-san91 Oh that would be a plus for me and generally more realistic.

@jamesjonnes Год назад

Yes, 6b is too horny for me as well. 13b seems less likely to do that from my experience.

@touma-san91 Год назад

@@LvZ90 If you want to run it, I hope you have atleast 12 gigs of VRAM. That's how much the quantized model needs. And it runs way faster using KoboldAI with the 4bit patch. Would give you a link how to do it but links are blocked here..

@zafiralpstv8004 Год назад

@@jamesjonnes it not depends on character that this Ai tried to play? if someone tag "nymphomaniac" than no wonder ai will be horny

@kahzy_fr Год назад

Does anyone know how to actually install this? I just get traceback errors from Ooogabooga whenever I try installing these

@sisyahonjourney Год назад

How can i get knowledge around this topic?

@dreamphoenix Год назад

Thank you.

@monkun76 Год назад

how do you even download this? The safetensors doesn't work for me

@christerwallentin1585 Год назад

Downloaded files from inside Oogabooga but I get phyton error that forces me to restart Oobabooga. Any ideas?

@Carrington1961 8 месяцев назад

Can someone explain why if context is above 1500 my token per second drops from 10 to 0.03?

@ia_para_Negocios 6 месяцев назад

how about requeriments for the pc ?

@yinny5601 Год назад

How can I use this for janitor Ai-? (Not from open Ai)

@zomgwtfbekjam Год назад

How about different languages like japanese? I'd love to have japanese conversations, because there's not too many japanese people I can talk to.

@shadowmasterp Год назад

I have not tried it yet but i seen a model trained on Japanese wiki. i think that would be your best bet.

Далее

Воссоздал СТРАНУ СМЕШАРИКОВ в Майнкрафт ХАРДКОР

25:47

Воссоздал СТРАНУ СМЕШАРИКОВ в Майнкрафт ХАРДКОР

Просмотров 293 тыс.

Бежать пока не поздно! Отец Андрей Ткачёв

00:59

Бежать пока не поздно! Отец Андрей Ткачёв

Просмотров 311 тыс.

PEDRO PEDRO INSIDEOUT

00:10

PEDRO PEDRO INSIDEOUT

Просмотров 3,2 млн

От первого лица: Лагерь 😱 УГНАЛИ ЯХТУ 🤯 РАЗГРОМИЛИ ЛАГЕРЬ 🥹 ВЫГНАЛИ из СТРАНЫ 😭 ГЛАЗАМИ ШКОЛЬНИКА

29:38

От первого лица: Лагерь 😱 УГНАЛИ ЯХТУ 🤯 РАЗГРОМИЛИ ЛАГЕРЬ 🥹 ВЫГНАЛИ из СТРАНЫ 😭 ГЛАЗАМИ ШКОЛЬНИКА

Просмотров 2,4 млн

Воссоздал СТРАНУ СМЕШАРИКОВ в Майнкрафт ХАРДКОР

25:47

Воссоздал СТРАНУ СМЕШАРИКОВ в Майнкрафт ХАРДКОР

Просмотров 293 тыс.