Do you know one you can run locally (preferably on budget hardware) that's worth using instead of going to chatgpt and just dealing with the restrictions? I've been thinking of setting one up on my dinky home server, but most of the time there limitations of my system, and the quality of available models I can find, make it not worth the effort.
There are some nicely optimized models using LLaMA as foundation (like Vicuna). The LLaMA derivatives can run on CPU even with llama.cpp
Otherwise you have to invest in a GPU, rule of thumb is number of parameters * 2 = VRAM required.
4 bit quantization can reduce the requirements quite a lot at little loss though.
If you want casual convo including adult content, OPT 6B derivatives are fairly good. Also, it's relatively easy to integrate some cross attention with BLIP or CLIP for multimodal understanding.
Don't care, there is more than enough alternatives which are sufficient at holding a casual convo. Best of all: You can run them locally.
Do you know one you can run locally (preferably on budget hardware) that's worth using instead of going to chatgpt and just dealing with the restrictions? I've been thinking of setting one up on my dinky home server, but most of the time there limitations of my system, and the quality of available models I can find, make it not worth the effort.
> can I compete with a corporation burning $700,000 per day to run an online chat bot?
I don't need to compete, it just needs to be barely good enough to be worth setting up.
stable diffusion seems to be doing okay. the quality of waifu it produces seems to only pale in comparison to midjourney but that's roughly it.
With a decent graphics card you can
/lmg/
No, you can't.
There are some nicely optimized models using LLaMA as foundation (like Vicuna). The LLaMA derivatives can run on CPU even with llama.cpp
Otherwise you have to invest in a GPU, rule of thumb is number of parameters * 2 = VRAM required.
4 bit quantization can reduce the requirements quite a lot at little loss though.
If you want casual convo including adult content, OPT 6B derivatives are fairly good. Also, it's relatively easy to integrate some cross attention with BLIP or CLIP for multimodal understanding.
> uses cloud services
> has no freedom
You don't say.
been there done that
how did you get terminated in 2021?
a bot in a 4chan IRC where all the users could talk with it. basically 20~ different people sending it racist things for a month via my API key.
zased
>organization Personal
Did ChatGPT write this? lol
Why do corporate psychopaths hate LLM sex so much?
Unfortunately it's only in response to the people who make a big deal about it on social media about it.
Why do Twitter trannies hate LLM sex?
because its another preferred type of sex that people have before they'd even consider sex with them?
why would this be a problem? doesn't the model just not fulfill such requests?