You mentioned how powerful your computer has to be. I'm currently running a ProArt Studio book with Intel i9 13980HX with a Nvidia 3000 ada GPU. Should this be powerful enough to run the larger models?
I do not recommend the largest for sure, my high end PC with 32GB ram and nice CPU couldn't answer a message in like 10 minutes. It maxed out my CPU and RAM as well
Haven't installed it on my machine yet (RTX 4090), but I doubt you could install the larger models on a 3000 very much, usually with quantized models I can run up to the ones with 32b params, I'm downloading the 70b one right now and crossing fingers, the largest model they have is 670b params (!!!), so an order of magnitude bigger, I guess those are exclusively for A100+ cards though.
It says I used tokens after asking my first question. I thought it was free - no limits. How many tokens do I get? Am I going to have to pay for more? Does the llama have more tokens?
i belive it prompted you with either selecting a blue or gray button to press. click the blue button then you should be able to follow his directions. hope that helps!
Why are you misleading viewers? the 70B model is NOT R1. Ollama even knows this. TH-camrs know this. the actual Model is 600B. the 70B is just the Qwen 2.5 Distilled version of DeepSeek.
Nice and short, not dragged out to 15 mins of video with needless banter. Thanks!
on point
Yep soon as I seen anything over 2 minutes. I exit the video.
Great vid bro, show us some of Deep seek image generation or anything about design generating
Confirmed working. And like others, really appreciate the sound & succinct presentation
You have all the 🍪🍪🍪🍪🍪
ine just shows ... for every response. im running on an m1 MacBook with 8gb ram. 1.5b model
Mine too idk what I did wrong
I am running the 14B version on a 16Gm M4 MacBook Air, making Sam Altman cry
Best video by far on the subject 👍🏼
I can almost heard jensen and their nvidia shareholder screaming in agony😂
Cool. I got it running on Linux Mint using this guide. I picked the 14 model, and it's slow on my machine, but it's working.
i installed it exactly this way on ubuntu, but I only get empty answers from the deepseek models (llama3 works fine). What is up?
How do I choose which drive to install to? I don't want it on my C drive.
You mentioned how powerful your computer has to be. I'm currently running a ProArt Studio book with Intel i9 13980HX with a Nvidia 3000 ada GPU. Should this be powerful enough to run the larger models?
I do not recommend the largest for sure, my high end PC with 32GB ram and nice CPU couldn't answer a message in like 10 minutes. It maxed out my CPU and RAM as well
@matrayzz lol I forgot the mention have 64gb of ram. Thank you! I'll stay away from the largest one. 😂
Haven't installed it on my machine yet (RTX 4090), but I doubt you could install the larger models on a 3000 very much, usually with quantized models I can run up to the ones with 32b params, I'm downloading the 70b one right now and crossing fingers, the largest model they have is 670b params (!!!), so an order of magnitude bigger, I guess those are exclusively for A100+ cards though.
@@matrayzz that stuff hardly matters, all about size of your vram
You should be fine running the 32b version
is it possible to deactivate the thinking part?
In the app it's called "Deep Think". Maybe ask it "Please turn off DeepThink until I ask for it again"
just delete the app
Im running the 32B R1 I want to create a voice activated one to act like siri for windows...
what's the requirement for 32b?
@@xlordfifth idk i have 32gb ram..
You wrote it first as if the (32) was a part of the version
@@ibrahem891 it is 32B means 32 Billion parameters, I read somewhere that 32B requires 32GB of ram to run...
@ It is part of the version, it's the 32Billion parameter version :/
will it search internet under such mode?
how do i uninstall the models?
ollama rm (reinput the model name)
@adrianwei5282 thanks
awesome video. you're a saint
What do the numbers mean ? How do I know which would be best for my pc ? Specs are 12600k, 32gig ram, 3080 +5700xt
14b will work fine, 32b will be slow.
It says I used tokens after asking my first question. I thought it was free - no limits. How many tokens do I get? Am I going to have to pay for more? Does the llama have more tokens?
probably just for resource management for their servers. To make sure people arent overloading it causing lags and crashes.
there are unlimited tokens, tokens refere to the lenth of your input/output of the AI
My Mac won't allow me to open Llama
Appreciate it brotha
It's not thinking or reasoning in real-time. It's predicting the next word
your thinking process is not far off
@ndjxisjenxjix9525 why are you gay
This is not for R1
Chatbox doesn't follow your easy example, wants stuff like API key, etc. license
i belive it prompted you with either selecting a blue or gray button to press. click the blue button then you should be able to follow his directions. hope that helps!
Thanks it works
Why are you misleading viewers? the 70B model is NOT R1. Ollama even knows this. TH-camrs know this. the actual Model is 600B. the 70B is just the Qwen 2.5 Distilled version of DeepSeek.
I didn’t understand
Ty
Attempting this on a NAS..! chatbot is super expensive.. pass.
Those aren't deepseek r1. They are just tuned down models that are interesting, that's all.
It works, but it is useless for coding... (8b)😑
what kind of coding? Is it wrong or does it just not know?
Can I use it for unit testing node js apps?
Dont use R1 for coding, use DeepSeek Coder.
@@Jamal-mq6xnIsn’t qwen 2.5 coder better?
Thanks a lot. Short and to the point 🫡❤️
World biggest controversial ai model 😂😂😂😂😂😂😂😅😅😅😅
th-cam.com/video/WGBz6XrGx_E/w-d-xo.htmlfeature=shared