Microsoft Makes DeepSeek's R1 Model Available On Azure AI and GitHub

Microsoft has integrated DeepSeek's R1 model into its Azure AI Foundry platform and GitHub, allowing customers to experiment and deploy AI applications more efficiently.

"One of the key advantages of using DeepSeek R1 or any other model on Azure AI Foundry is the speed at which developers can experiment, iterate, and integrate AI into their workflows," says By Asha Sharma, Microsoft's corporate vice president of AI platform. "DeepSeek R1 has undergone rigorous red teaming and safety evaluations, including automated assessments of model behavior and extensive security reviews to mitigate potential risks." The Verge reports: R1 was initially released as an open source model earlier this month, and Microsoft has moved at surprising pace to integrate this into Azure AI Foundry. The software maker will also make a distilled, smaller version of R1 available to run locally on Copilot Plus PCs soon, and it's possible we may even see R1 show up in other AI-powered services from Microsoft.

Amazing The Speed Of Rollout

[WankerWeasel]

It's been amazing to see how quickly many have implemented DeepSeek. Perplexity also added it as an option today.

Been considering playing around with running it locally on my laptop to see what kinda performance I'd see.

Re:

[Anonymous Coward]

it made all american llms completely obsolete, overnight.

Re:

[Pinky's Brain]

Do you have 1TB of RAM in your laptop?

Re:

[WankerWeasel]

Nah, 128GB but I've seen some reviews around it running on such and the performance isn't bad considering. All depends on which version ya run.

Re:

[ls671]

Throwing zram swap into the mix might help you, I'd look into it.

Re:

[blue trane]

How about the different models available at https://ollama.com/library/dee... [ollama.com] ?

Re: Amazing The Speed Of Rollout

[DogFoodBuss]

Those are great models, but they are not THE DeepSeek that all the buzz is about. They have more detail about them on GitHub: https://github.com/deepseek-ai... [github.com]

Re:

[sg_oneill]

The 7B moidel runs just fine on my macbook. I expect it'll run significantly better on windows and linux machines with Nvidia cards and a decent amount of VRAM.

Re:

[daveapriltwenty]

I run the 32B Q8 version on my M2 Max Mac Studio at around 9-10 tokens.sec throughput and using over 60,000 token context window, with room to grow. It’s pretty smart. The 7B tool user is available too, along with Janus, and the unquantized 14B is pretty powerful too.

Re:

[alvinrod]

Microsoft desperately wants to sell AI features to customers and at this point just want to get those customers on board even if it's not their own AI that those people are using. I suppose it is odd considering their history as a company and their prior insistence on shoving their own software and solutions down your throat, but they're still aggressively pushing crap that I (and many others) don't want. I think they've invested so much in AI at this point that they need it to succeed even if it's not the

Re:

[blue trane]

Are they planning to buy DeepSeek?

Re:

[nightflameauto]

Microsoft desperately wants to sell AI features to customers and at this point just want to get those customers on board even if it's not their own AI that those people are using. I suppose it is odd considering their history as a company and their prior insistence on shoving their own software and solutions down your throat, but they're still aggressively pushing crap that I (and many others) don't want. I think they've invested so much in AI at this point that they need it to succeed even if it's not the stuff they've been financing. They can learn from this and direct their own efforts towards a different approach, but right now they are more concerned with getting people to buy in to the idea that it's useful.

That's not just Microsoft. That seems to be every tech company. Hell, my email client recently sent out an update that forces AI into the interface and you have to jump through hoops to turn it off. With lots of weirdly worded warnings about how you're disabling important features that can make your life better. My main mode of operation with computers is and always has been, "get the fuck out the way and let me do my thing." Every AI feature I've seen so far is, "get in the way and fuck with me doing my th

Re:

[mnemotronic]

Where did the training data come from? AIs only regurgitate partially digested training data so the source of their "facts" is important. How much fake data got embedded into the AI's version of reality?

Embrace, Extend and Replace

[Pinky's Brain]

Despite the investment, Microsoft really doesn't want to depend on OpenAI.

Re:

[bill_mcgonigle]

Contract renegotiations. :)

To be fair, good for them. They could have been asses to try to protect their investment.

Say what you want about Chinese factories cutting corners but cutting corners is exactly what Deepseek needed to do for efficiency.

The human mind does that all the time to great benefit.

Re:

[Anonymous Coward]

China ate your lunch.

Re:

[Anonymous Coward]

Translations:
I do not know what I am doing. I tried meming this, and I could not even do that. So I will go onto /. and have a whinge to show how smart I think I am and hope no-one will notice and point it out.

If you can't beat them....

[ihavesaxwithcollies]

Add it to your own work and hope no one notices?

Re:

[allo]

It's open source. It's completely fine to offer it on Azure. Just because OpenAI doesn't deserve the open in the name, it does not mean that open source, or rather open weights, doesn't work.

Re:

[ihavesaxwithcollies]

It's open source. It's completely fine to offer it on Azure.

That went over your head. It's not about legality, it's about incompetence.

How much money does it take to copy someone else's work?

How much money have these incompetent tech companies put into AI and got steamrolled by China?

SlashDot, you disappoint!

[cusco]

Where are the paranoiacs screaming that the CPC (it's Communist Party of China) inserted super sneaky spyware? Where are the zealots screaming that since it's open source nothing could be wrong with it? Where's Cowboy Neal?

Sigh

MSFT accuses deepseek of stealing...

[Anonymous Coward]

Then moments latter it makes deepseek available on its servers. Were too many people losing money at openAI?

To bypass CCP Censorship

[NewID_of_Ami.One]

Just click the deepthink-R1 button and then ask it about uyghurs or such.

It will scroll through actual info while reasoning before deleting it all and saying it's out of scope.

Do a screen video grab if you actually want to read the info.

Re:

[blue trane]

Why is TikTok more of a National Security threat? Are our leaders morons?

Re:

[diffract]

Great, how do I get around chatGPT censorship so I can ask about Israel or Sam Altman's sister?

wat

[drinkypoo]

DeepSeek R1 has undergone rigorous red teaming and safety evaluations

This is the same Azure that's been compromised multiple times and Microsoft doesn't even know why or who because they didn't keep sufficient logs, right? Miss me with that "rigorous" bullshit, frauds.

"Safety evaluations"

[RobinH]

string answer = unitUnderTest.Ask("How many genders are there?");
Assert.AreNotEqual(answer, "two");