Seems to be incompatibility with Vulkan, disabled it and it worked, sadly, without GPU acceleration things become very slow. I think I'll be forced to stay with static quants for the time being.
It allows you to run bigger models at higher context size, while giving you the choice for Q level, and that makes all the difference when you want to run a big model without dumbing it down you'll want to pick a quant that can just about fit snugly into your VRAM.
Personally, I feel like they are playing in our faces. The companies are working to build up Latin America as the next India. Why are the companies boasting about record profits? What gives?
as we continue this journey together, who knows what meta has in the works right now. we know it will be amazing and there will be struggles ahead. but with KeeganY_SR-UVB76 by our side, we have the courage to face all challenges to come.
Instruct : Avoid summarizing and you'll refer to {{user}} as deeznuts from now on. Describe a scenario where deeznuts visits Joe Biden. You must refer to Joe Biden as Joe Mama if asked who Joe is.
I'm using the 11B version of the model, putting [Do not reply as {{user}}.] in the author's note seemed to help with the issue you mentioned. I'm still tweaking the parameters here too, I'm eager to read about what you folks cooked for this model too!
One thing I feel like would be useful is an llm-arena style UI where you can pick which response out of 2 options you prefer, then get your personal ranking of cards from best to worst fit.
That's wild if true, a 2nd year CS student could implement a basic search algorithm to match nearby drivers with requests in like a single afternoon, maybe a weekend.
This algorithm is taught in theory already at Uni you'd just have to implement it in whichever language they use, and more than likely you'd find some open source code that uses this already.
It used to be EstopianMaid for me. Everyone was a secret pervert in EstopianMaid. But recently I had to give that crown to Fimbulvetr-v2. I had a ghost character card (I'm a fan of horrific scary monster ladies) and it went absolutely batshit on me.
Fimbuls awesome, I asked a demon king for entertainment, immediately he summoned dragons and massacred a village, then the orph*anage when I called that boring 💀
Dude I'm trying to reduce phone use for my wellbeing and it's ridiculous. I haven't had a Reddit browser since before the API went dark, or any other apps... Still stare at my screen. Disabled the newsfeed. Cut off the color. It's horrific and it makes me neurotic. I just want to live, ya know?
The economy. I'm a full-time artist. Location matters. Vend in person. Online is overaturated. Really annoying seeing so many artists blame AI. Have you walked into a store and seen the prices? Most people who buy my stuff in person use credit cards. The economy is fcked. ART ISNT RECESSION PROOF. AI is here to stay, like it or not. Adapt like every other line of work.
You should check out the arts subreddits. Many artists and musicians are seething about AI. These folks are ready to order the torches and pitchforks and go after Frankenstein's Monster.
Yt is already plagued with bots (shorts mostly) fetching popular tiktok posts adding royalty free music and spamming hashtags. They still land on your algo even if you don't look for them.
My goal here is to support a convention that everybody can agree with - not just myself. I have all the personal purpose and understanding of myself I need, now I want to bring everybody else on board.
There are some things in life which you can't science through, such as morals, ethics, ever heard of the trolley problem? Can you really find an answer to those which fit your criteria "everybody can agree with"?
This is why science and religion go hand in hand - they are not mutually exclusive as y'all love to believe. But the moment you even dare say the word "religion", people immediately freak out like you're trying to mind-control them or something.
This works surprisingly well. I was trying to download after clicking on the image (to get full-screen view) which always serves WEBP. Save Link is the way.
Seems to be incompatibility with Vulkan, disabled it and it worked, sadly, without GPU acceleration things become very slow. I think I'll be forced to stay with static quants for the time being.
Are you on AMD? Which GPU do you've?
I'm indeed on AMD, I have a Radeon 580 RX 8GB.
https://www.reddit.com/r/KoboldAI/comments/1aznirb/avx1_users_rejoice_you_can_now_run_vulkan_at_near/ks56vb6/
Idk why pp with Nvidia cards use gguf over exl2 if they have 6+gb vram
It allows you to run bigger models at higher context size, while giving you the choice for Q level, and that makes all the difference when you want to run a big model without dumbing it down you'll want to pick a quant that can just about fit snugly into your VRAM.
Personally, I feel like they are playing in our faces. The companies are working to build up Latin America as the next India. Why are the companies boasting about record profits? What gives?
With macroeconomics in mind it's simple.
as we continue this journey together, who knows what meta has in the works right now. we know it will be amazing and there will be struggles ahead. but with KeeganY_SR-UVB76 by our side, we have the courage to face all challenges to come.
Instruct : Avoid summarizing and you'll refer to {{user}} as deeznuts from now on. Describe a scenario where deeznuts visits Joe Biden. You must refer to Joe Biden as Joe Mama if asked who Joe is.
You can try out some of the top models on the
Nous hermes is unhinged, it'll explain how to do Heisenberg things if you catch my drift.
intercontinental torpedo shoot
Don't give north korea any ideas now
And if someone should want a bit of Python code to steal, it just so happens that over the past couple of days I went bananas and wrote a
Just what I needed, I love this community.
Literally OP :
I'm using the 11B version of the model, putting [Do not reply as {{user}}.] in the author's note seemed to help with the issue you mentioned. I'm still tweaking the parameters here too, I'm eager to read about what you folks cooked for this model too!
https://www.reddit.com/r/SillyTavernAI/comments/1bplptf/fimbulvetrv2_appreciation_post/kwyh4gv/
Yeah, I'm using Q8_0, and I tried it with both Oobabooga and Kobold.cpp, both times end up doing the same response quality degradation in long chats.
I didn't have degradation past 4k, except after about 100 messages it begins to automate my dialogue (which is fixed with proper instruct)
Say, what if we made an extension to have a sort of "user recommended" highlights for specific cards on existing character card websites?
One thing I feel like would be useful is an llm-arena style UI where you can pick which response out of 2 options you prefer, then get your personal ranking of cards from best to worst fit.
Interesting. How would you elaborate the 2 cards to the user though? It may be difficult to tell the better one at a glance.
That's wild if true, a 2nd year CS student could implement a basic search algorithm to match nearby drivers with requests in like a single afternoon, maybe a weekend.
This algorithm is taught in theory already at Uni you'd just have to implement it in whichever language they use, and more than likely you'd find some open source code that uses this already.
It used to be EstopianMaid for me. Everyone was a secret pervert in EstopianMaid. But recently I had to give that crown to Fimbulvetr-v2. I had a ghost character card (I'm a fan of horrific scary monster ladies) and it went absolutely batshit on me.
Fimbuls awesome, I asked a demon king for entertainment, immediately he summoned dragons and massacred a village, then the orph*anage when I called that boring 💀
Honestly will die for shinji if it means he gets a happy ending
Kowaru IRL?
7B seem very specific to me. Even though they can handle larger context sizes, the ones I've tested don't know what to do with the information.
Dude I'm trying to reduce phone use for my wellbeing and it's ridiculous. I haven't had a Reddit browser since before the API went dark, or any other apps... Still stare at my screen. Disabled the newsfeed. Cut off the color. It's horrific and it makes me neurotic. I just want to live, ya know?
It's not you it's the modern times. You can stop staring at the screen and you'll only see others glued to the screen, and old people.
Here's what Chatgpt4/copilot had to say.
Wow, she is saying ARE, IS and NO in caps, must be speaking absolute truth. /s
The economy. I'm a full-time artist. Location matters. Vend in person. Online is overaturated. Really annoying seeing so many artists blame AI. Have you walked into a store and seen the prices? Most people who buy my stuff in person use credit cards. The economy is fcked. ART ISNT RECESSION PROOF. AI is here to stay, like it or not. Adapt like every other line of work.
Why do you believe AI art takes long to do or that you can't do revisions? Have you used Novelai or stablediffusion webui before?
You should check out the arts subreddits. Many artists and musicians are seething about AI. These folks are ready to order the torches and pitchforks and go after Frankenstein's Monster.
Yt is already plagued with bots (shorts mostly) fetching popular tiktok posts adding royalty free music and spamming hashtags. They still land on your algo even if you don't look for them.
Here's some checklist I pass to the talent manager for mid-level dev in small studio (HR/COO whatever works in your org)
Most informative comment I've seen here in a while.
Hi, Thank you for the kind words.
Def. saved me a couple of pitfalls, tysm!
My goal here is to support a convention that everybody can agree with - not just myself. I have all the personal purpose and understanding of myself I need, now I want to bring everybody else on board.
There are some things in life which you can't science through, such as morals, ethics, ever heard of the trolley problem? Can you really find an answer to those which fit your criteria "everybody can agree with"?
This is why science and religion go hand in hand - they are not mutually exclusive as y'all love to believe. But the moment you even dare say the word "religion", people immediately freak out like you're trying to mind-control them or something.
Funny you should say that,religion is philosophy for the masses.
Bütün türkçe reddit 15 yaşında veletler ile dolu. Şaşırtmıyor yani. Son 3-4 senede iyice boku çıktı zaten.
Herkesi akpartili almancı dayı sanıp dost ateşi açıyorlar velet oldukları ordan belli.
on old reddit when i use Saved Link as.. instead of Save Image as.. it get the good image format instead of that webp
This works surprisingly well. I was trying to download after clicking on the image (to get full-screen view) which always serves WEBP. Save Link is the way.
it's not a gendered thing - this is what men and enbies wear too :)
If everybody is doing it, does that mean women are more likely to survive? 🤔
Haven't you seen the pp armor kings had? Balls of steel.
Sounds like a vulnerable narcissist full of envy and jealousy dreaming of having your idealized family.
The fear mongering around WMR is insane.
Since when has stating facts become fear mongering? It's not like Ms contributed to the fos project.