News DeepSeek.com making waves!

vishalrao

vishalrao

Global Moral Police
Keymaster
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!
 
  • Like
Reactions: roadrippersid, Skoomahed and lockhrt999
vishalrao said:
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!
Click to expand...
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.
 
  • Like
Reactions: vishalrao
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).

1737817335097.webp


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]
https://www.reddit.com/r/LocalLLaMA/comments/1i80cwf/deepseek_is_a_side_project
 
  • Wow
  • Love
Reactions: vishalrao and peanutbutter
Skoomahed said:
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.
Click to expand...

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 :laughing:
 
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.
 
  • Like
Reactions: Pat and ibose
Top Bottom