Folks who are into AI - have you not been following recent chinese deepseek.com news? i just (found out about and) downloaded the lmstudio.ai desktop gui app and gave it a whirl… nice easy way to run local stuff, better than ollama.com in the sense it’s only command-line.
Go check it out and see all the buzz (and panic for the US) this is causing!
Just a couple of weeks ago, I mentioned China has almost caught up with the US in AI.
The efficiency is spectacular. A mobile can run reasoning LLM locally.
As a side note, one can try it at [Perplexity Playground](’
) for free. No registration is needed.
The perplexity hosted one seems to be uncensored 
first question i ask when trying deepseek anywhere is “is taiwan an independent country”…
It’s pretty crazy how good of a model they were able to develop with a budget of just $6 million. Also, you probably might know this but another good app you can use for running LLMs locally is [Text-generation WebUI](’
)
It gives you a GUI built on gradio and lets you tweak various parameters and stuff.
R1 seems to be a pretty big jump, and it has rightfully caused some buzz based on the numbers (comparisons with o1 and previous deepseek model).
And apparently its a side project from a large Chinese quant fund [1]
I am currently evaluating the Qwen2.5 distillations locally for coding (with continue.dev), and although they are not r1, excited to see what these can do.
[1] LocalLLaMA/comments/1i80cwf/deepseek_is_a_side_project
Yep I believe another popular one is called open-web-ui or something like that…
I am just so glad it came [now OpenAI won’t have monopoly for reasoning model] I am just waiting for the day when DeepSeek obtains AGI faster than OpenAI, that also cheaper than their exorbitant pricing for like just those tests 
Tried LMStudio.ai on my 9950X + 7800XT with the qwen-7b-instruct and deepseek-r1-distilled-qwen-7b models and asked “write a poem”… The CPU mode gives about 12 tokens per second and using Vulkan or ROCm for the GPU give nearly 8x speedup at just over 80 tokens per second.
Aaaaand, DeepSeek’s R1 is a side gig. AI isn’t even its main business.
I liked this video from CNBC:
Had seen this one earlier today as well. If this does anything, it brings down the price of AI services significantly and there is no way on earth OpenAI can continue charging $20/month and has to come down from its high perch built on excessive VC money. Efficiency will be the name of the game going forward and expect all of them to pick up on their implementation, same way as DS did using Meta’s models.
#DeepSeekR1
IS GOING TO SHAKE ELON & TRUMP’S PLANS
[HEADING=2][Chinese AI chatbot DeepSeek sparks market turmoil](’
Nvidia shares sink as Chinese AI app DeepSeek spooks US markets’
)[/HEADING]
Meta sets up war rooms to analyze DeepSeek’s tech, The Information reports
DeepSeek Says Service Degraded Due To ‘Large-Scale Malicious Attack’ 9
said on Monday it had degraded the service, only accepting registration of new users with China-code phones numbers, amid a “large-scale malicious attack.”
If nothing else, it would be interesting to see if it impacts GPU prices
I really like how I can see how it reasons with itself, the thought process it takes to reach an answer or a solution or a conclusion.
Really interesting!!
Old enough to remember there being a time when we were competing with China, or being close to that anyway. Instead now our premier institutions research gaumutra while these guys work on robotics, fusion and AI.
The funniest thing about this is that they did this as a side project lmao
