Deepseek R1-0528

Well that didn&#x27;t take long, available from 7 providers through openrouter.<p><a href="https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-r1-0528&#x2F;providers" rel="nofollow">https:&#x2F;&#x2F;openrouter.ai&#x2F;deepseek&#x2F;deepseek-r1-0528&#x2F;providers</a><p>May 28th update to the original DeepSeek R1 Performance on par with OpenAI o1, but open-sourced and with fully open reasoning tokens. It&#x27;s 671B parameters in size, with 37B active in an inference pass.<p>Fully open-source model.

No information to be found about it. Hopefully we get benchmarks soon. Reminds me of the days when Mistral would just tweet a torrent magnet link

I love how Deepseek just casually drops new updates (that deliver big improvements) without fanfare.

Out of sheer curiosity: What’s required for the average Joe to use this, even at a glacial pace, in terms of hardware? Or is it even possible without using smart person magic to append enchanted numbers and make it smaller for us masses?

For those interested, I made some 1 bit dynamic quants at <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-R1-0528-GGUF" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;unsloth&#x2F;DeepSeek-R1-0528-GGUF</a><p>74% smaller 713GB to 185GB.<p>Use the magic incantation -ot &quot;.ffn_.*_exps.=CPU&quot; to offload MoE layers to RAM, allowing non MoEs to fit &lt; 24GB VRAM on 16K context! The rest sits in RAM &amp; disk.

What use cases are people using local LLMs for? Have you created any practical tools that actually increase your efficiency? I&#x27;ve been experimenting a bit but find it hard to get inspiration for useful applications

Not much to go off of here. I think the latest R1 release should be exciting. 685B parameters. No model card. Release notes? Changes? Context window? The original R1 has impressive output but really burns tokens to get there. Can&#x27;t wait to learn more!

I think it’s cool to see this kind of international participation in fierce tech competition. It’s exciting. It’s what I think capitalism should be.<p>This whole “building moats” and buying competitors fascination in the US has gotten boring, obvious and dull. The world benefits when companies struggle to be the best.

Deepseek seems to be one of the few LLMs that run on a iPod Touch because of the older version of ios.

671B parameters! Well, it doesn&#x27;t look like I&#x27;ll be running that locally.

You&#x27;re gonna need at least 8 h100 80s for this....

About half the price of o4 mini high for not that much worse performance, interesting<p>edit: most providers are offering a quantized version...

I want to see it in groq asap !