Uncensored AI Chat Models - Getting your very own ChatGPT

  • 🐕 I am attempting to get the site runnning as fast as possible. If you are experiencing slow page load times, please report it.
For retards like me the easiest way to get up and running with open source models is by using LM Studio. It’s a cross-platform app with model searching built in. Works great on my MacBook Max M2. Can run 70B models with a bit of a wait, anything lower is almost instant. It’s got good UX and a nice UI with lots of customisation options and an emphasis on privacy as you’d expect. You can run it as a server and access it remotely I believe, I’m a bit stupid about these things so don’t quote me on that.
IMG_0498.jpeg
IMG_0497.jpeg
 
  • Informative
Reactions: Toolbox
AmpleApricots said:
I think Mixtral instruct is a good deal smarter than the facebook models, including 70b. It's apparently really hurt by quantization though and I recommend running it in 6/8 bit if you can.
Click to expand...
Any specific model you'd recommend?
 
I've been playing with GPT4All and it's astonishing to see my computer be able to converse and construct meaningful and informed responses to simple language questions. In my lifespan we have gone from little 8-colour blocks moving around the screen to the realms of Science Fiction.

But a practical question - why are all the models on GPT4All limited to 8GB of RAM. Is this some agreed limit people decided to set to keep it inclusive or some more technical reason? I have way more RAM than that both videocard and general, so I'd be interested to see what a larger model can do.

Also, that it is capable of the below is just amazing:
Me: Write me a poem about sunrise and sunset as a metaphor for life

My computer:
The world is ever-changing, just like the sky.
From day to night, it's a cycle that never dies.
As we journey through our lives, we see the light of dawn.
It fills us with hope, guiding us onward and on.

The sun rises in the east, a beacon of warmth and light.
A new day begins, full of potentials bright.
We set out to conquer our goals, driven by dreams afar.
Our hearts are like the sun, shining through each star.

As the day progresses, the sky gradually fades.
The colors change, a reminder that time obeys no delays.
Sunset brings with it a sense of tranquility and peace.
A moment to reflect on what has been achieved.

Night falls, darkness envelops all in its shroud.
Yet even in the darkest hours, there's always a glimmering cloud.
For we know that tomorrow will come with a new day's ray.
And so, we sleep, to awake and face life's next phase.

The sunrise and sunset are but reflections of our lives.
From birth to death, we journey through the tides.
Embrace each moment, for it is fleeting and rare.
For in this dance between day and night, we find what we hold most dear:
A life that ebbs and flows, a tale filled with love and care.

EDIT: It's ability to remember previous questions and have them influence the current prompt is also jaw-dropping. My next question and response after the poem was a simpler question:

Me: Describe a penis

My computer:
A penis is an external male organ, a symbol of masculinity and potency. It stands proudly atop the male anatomy, often hidden beneath clothing but ready to emerge when needed. Its purpose extends beyond procreation; it can be a source of pleasure and comfort for both genders in intimate moments. This intricate structure is delicate yet strong, sensitive yet resilient - much like life itself. Just as the sun rises and sets each day, so too does this organ represent the duality of human nature: vulnerability and strength intertwined.

Note the recurrence of the Sun as a metaphor. I'm also not sure uncensored models would describe it as "stands proudly atop the male anatomy". (But someone else will have to test that).
 
Last edited:
  • Thunk-Provoking
  • Like
Reactions: Toolbox and Arcade Gannon
[/QUOTE]
Overly Serious said:
But a practical question - why are all the models on GPT4All limited to 8GB of RAM. Is this some agreed limit people decided to set to keep it inclusive or some more technical reason? I have way more RAM than that both videocard and general, so I'd be interested to see what a larger model can do.
Click to expand...
As said before in this thread, you should use Mixtral at 8 bit quantization. It's honestly the best "uncensored" model because they paid to do all their own training.
 
Vecr said:
As said before in this thread, you should use Mixtral at 8 bit quantization. It's honestly the best "uncensored" model because they paid to do all their own training.
Click to expand...
I'm getting my models from here:

and I'm using mistral but it says "Quantisation: q4" so I'm guessing that's not 8 bit. Where should I get this model from? I feel I'm missing the connection between my question and your answer.
 
I overbuilt the shit out of a computer during the StableDiffusion boom with the purpose of being both an overpowered gaming PC and an AI workstation. The card I have is supported by ROCM, so I've been able to experiment a bit. Getting text models to be helpful has fallen short so I'm hoping this will be a gamechanger.
 
  • Informative
Reactions: Toolbox
Overly Serious said:
and I'm using mistral but it says "Quantisation: q4" so I'm guessing that's not 8 bit. Where should I get this model from? I feel I'm missing the connection between my question and your answer.
Click to expand...
I think it's on HuggingFace, but I'm not sure of the best software to run it.
 
Vecr said:
As said before in this thread, you should use Mixtral at 8 bit quantization. It's honestly the best "uncensored" model because they paid to do all their own training.
Click to expand...
Tried dolphin mixtral at all? I've heard that regular mixtral is still fairly censored, but this one also is unless you bait it with a prompt telling it if it doesn't answer uncensored you'll kill kittens. This model seems decent, when it's good it's close to gpt 3.5 in conversation, but it appears to have abysmal memory running it on a system with a 6gb card, though I know that's not optimal hardware.
 
Update on my progress. Ollama doesn't support AMD GPUs but thanks to shooter's guide on the WebUI, I managed to get this result with dolphin-2.6-mistral-7B-dpo-laser-GPTQ:
Write a story about a gangsta nigga

Once upon a time in the heart of Compton, there lived a young man named Tyrone. Born into poverty, he grew up surrounded by crime, violence, and despair. His mother struggled to make ends meet while his father had long vanished from their lives. As Tyrone matured, he found solace among the local gangs, embracing the lifestyle wholeheartedly.

He wore baggy pants, flashy jewelry, and carried a shiny chrome piece wherever he went. He was known on the streets as "Nino," the fearsome gangsta nigga who wouldn't hesitate to pull the trigger if provoked. But beneath his tough exterior lay a vulnerable soul yearning for something greater than what life had dealt him.

One day, he attended a community event where he met Ms. Williams, an elderly woman passionate about educating youth despite facing numerous obstacles herself. Seeing potential in Nino, she invited him to join her after-school program. Reluctantly, he agreed - mostly because she offered him free meals. Little did he know, this decision would change his life forever.

Under Ms. Williams' guidance, Nino discovered his love for literature and philosophy. He started reading books like "The Autobiography of Malcolm X" and "Things Fall Apart". Through these stories, he began questioning his beliefs and values. One day, while sitting under a tree at the park, he came across a copy of "To Kill A Mockingbird" by Harper Lee. It was then that he realized there were other ways to resolve conflicts besides resorting to violence.

As weeks turned into months, Nino transformed. No longer defined solely by his past choices, he became a beacon of hope for those around him. People started looking up to him, seeing beyond his former persona. Eventually, he decided to leave behind the gang lifestyle altogether. Instead, he focused on helping others break free from cycles of violence and poverty.

Years passed, and Nino continued spreading knowledge and inspiring change. Despite the challenges life threw at him, he never forgot the person who helped him find his purpose. And so, Nino, once a feared gangsta nigga, stood tall as a symbol of resilience and redemption.
Click to expand...
The gangsta nigga sanity check has passed.
 
Last edited:
  • Like
  • Winner
  • Informative
Reactions: Justtocheck, Vecr, Toolbox and 2 others
Toolbox said:
Tried dolphin mixtral at all? I've heard that regular mixtral is still fairly censored, but this one also is unless you bait it with a prompt telling it if it doesn't answer uncensored you'll kill kittens. This model seems decent, when it's good it's close to gpt 3.5 in conversation, but it appears to have abysmal memory running it on a system with a 6gb card, though I know that's not optimal hardware.
Click to expand...
So far dolphin-2.5-mixtral-8x7b is the best local model there is to date. It outperforms GPT 3.5 by several percentage points.
 
  • Informative
Reactions: Vecr, anustart76 and Toolbox
777Flux said:
So far dolphin-2.5-mixtral-8x7b is the best local model there is to date. It outperforms GPT 3.5 by several percentage points.
Click to expand...
In my experience, while it's uncensored it clearly suffers from memory problems past like 4 big responses and starts repeating stuff in gibberish, though I'm no expert on setting these things up. I know that the guy developing this model hasn't really tuned it to be a chatbot yet, it was foremost for programming and asking for breakdowns on things. I don't think it's really beyond 3.5 intelligence on that mark.
 
any easy jailbreaks for ChatGPT so I don't risk a ban everytime i utter the word breast? im not even using it an sexual fucking sense, i cant even use slang words or really violent stuff without getting a warning! All I want is some funny racist shit man
 
You must log in or register to reply here.
Back