Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Kimi-K3 on HuggingFace

huggingface.co

Post reply on HN
Print view View original

581–588 of 588 posts

Re: Kimi-K3 on HuggingFace

#581

Post by sam345 »

nullbio wrote: post #551
ataoz wrote: post #530

I asked "Tell me about yourself" on HF. This is the response... Curious. > Kimi K3: I'm Claude, an AI assistant created by Anthropic. I'm built to be helpful with a wide range of tasks—things like writing and editing, answering questions, coding, analysis, brainstorming, explaining concepts, math, and creative projects.

This really means nothing. You can ask the same thing in Chinese to Opus and it will tell you it's DeepSeek, for example. Things leak into the training data and models hallucinate.

I don't think so.

Re: Kimi-K3 on HuggingFace

#582

Post by Tepix »

bjackman wrote: post #460
Tepix wrote: post #214

Earlier quoted context omitted.

"Never" is a long time. Just think about how much ram we had 10 or 20 years ago. 1.5TB isn't a lot really.

It doesn't matter if you have the RAM, running a 1.5TB model for a single context stream is fundamentally inefficient.

Lots of things are inefficient in IT, yet we do them anyway. Think about the amount of time your laptop is idle.

Re: Kimi-K3 on HuggingFace

#583

Post by mindslight »

rootusrootus wrote: post #535
mindslight wrote: post #525

Earlier quoted context omitted.

Then yes I'd agree with you, it doesn't make sense to describe your EV charging as costing $0.09/kWh since you're presumably only on the slightly more expensive ToD plan due to the EV. Personally I'd love to have a ToD plan, especially with the rate structure you've laid out - break even seems to be using less than 1/3 of your daily electricity usage during the on-peak hours, which is only 1/6 of the day? I've got a…

> FWIW I'd think the pricing of the ToD plan versus the fixed rate plan has more to do with how the power company themselves has to buy power and model/hedge against demand at various parts of the day, rather than simply trying to make the costs even for the consumer. I agree, my single point of evidence to support my theory is that the power company advertises the TOD plan by presenting an average customer with thei…

They might just not want to encourage people to switch too hard, lest people blindly sign up for it thinking big savings, their bills spike up, and the poco gets a huge customer service problem. Instead they show just a few pennies savings, and lure only the proactive type of people whose start thinking "... and I could save even more by adjusting my usage" ? Just a thought.

Re: Kimi-K3 on HuggingFace

#584

Post by naveen99 »

mrfox321 wrote: post #249
naveen99 wrote: post #244

Earlier quoted context omitted.

Isn’t that distillation ?

No. Distillation trains on a teach model's logits or output tokens.

How do you have teacher model output anything without it being output tokens (or embeddings / intermediate logits / activations)?

Re: Kimi-K3 on HuggingFace

#585

Post by cmrdporcupine »

trollbridge wrote: post #518
cmrdporcupine wrote: post #371

Earlier quoted context omitted.

> There will be a huge market for local inference once it's cheap and widely available. I've seen public pronouncements that the RAM shortage could persist for a decade. And then if consider that the constraint on local LLMs isn't just memory size but bandwidth ... If you take something like a DGX Spark and increase its memory to 512GB that doesn't even solve the problem. Because the bandwidth of DDR5 just can't mana…

Isn’t that innovation precisely what Qwen did?

They are one of many working on this. It's obviously the holy grail.

Re: Kimi-K3 on HuggingFace

#586

Post by pmxi »

wren6991 wrote: post #419
pmxi wrote: post #384

Earlier quoted context omitted.

That's a summarized and filtered view of the actual reasoning. OpenAI and Anthropic guard the real reasoning closely. Users have never been able to see it and the API returns an encrypted blob instead of legible reasoning.

Older models did show the full unredacted thinking trace, but I don't think Opus has ever shown full CoT. Here is an archived version of Anthropic's API docs saying that Sonnet 3.7 (only) has unredacted CoT on API: https://web.archive.org/web/20260324051339/https://platform....

That’s cool. I knew o1 hid it since launch, so I assumed Anthropic would also have never shown it.

Re: Kimi-K3 on HuggingFace

#588

Post by fnordpiglet »

Aboutplants wrote: post #481
nancyminusone wrote: post #456

Earlier quoted context omitted.

they will just restrict you from buying the hardware these run on

The rising price of hardware is already doing just that

Economics says the hardware required and the scale of hardware available at what price will converge to meet demand
Post reply on HN