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

Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

twitter.com

Post reply on HN
Print view View original

111–120 of 194 posts

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#111

Post by Alifatisk »

alt187 wrote: post #107
satvikpendem wrote: post #75

This is not open source, they have a "modified MIT license" where they have other restrictions on users over a certain threshold. Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you…

What part of this goes against the four fundamental freedoms? Can you point at it?

Exactly, I wouldn’t mind adding that text on our service if we made 20m $, the parent made it sound like a huge clause

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#112

Post by neuroelectron »

gpm wrote: post #93
summarity wrote: post #64

Earlier quoted context omitted.

Reasonable speeds are possible with 4bit quants on 2 512GB Mac Studios (MLX TB4 Ring - see https://x.com/awnihannun/status/1943723599971443134 ) or even a single socket Epyc system with >1TB of RAM (about the same real world memory throughput as the M Ultra). So $20k-ish to play with it. For real-world speeds though yeah, you'd need serious hardware. This is more of a "deploy your own stamp" model, less a "local" mod…

> or even a single socket Epyc system with >1TB of RAM How many tokens/second would this likely achieve?

1

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#113

Post by Alifatisk »

brcmthrowaway wrote: post #56

Is Kimi the new deep seek?

It kinda feels like it, but Moonshots delivery has been like this before aswell, it was just now their new release got way more highlight than usual. When they released Kimi k1.5, those bench were impressive at the time! But everyone was busy with Deepseek v3 and QwQ-32B

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#115

Post by simonw »

alt187 wrote: post #107
satvikpendem wrote: post #75

This is not open source, they have a "modified MIT license" where they have other restrictions on users over a certain threshold. Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you…

What part of this goes against the four fundamental freedoms? Can you point at it?

"The freedom to run the program as you wish, for any purpose (freedom 0)."

Being required to display branding in that way contradicts "run the program as you wish".

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#117

Post by Alifatisk »

ozgune wrote: post #57

This is a very impressive general purpose LLM (GPT 4o, DeepSeek-V3 family). It’s also open source. I think it hasn’t received much attention because the frontier shifted to reasoning and multi-modal AI models. In accuracy benchmarks, all the top models are reasoning ones: https://artificialanalysis.ai/ If someone took Kimi k2 and trained a reasoning model with it, I’d be curious how that model performs.

Why hasn’t Kimis current and older models been benchmarked and added to Artificial analysis yet?

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#118

Post by simonw »

drawnwren wrote: post #109
satvikpendem wrote: post #75

This is not open source, they have a "modified MIT license" where they have other restrictions on users over a certain threshold. Our only modification part is that, if the Software (or any derivative works thereof) is used for any of your commercial products or services that have more than 100 million monthly active users, or more than 20 million US dollars (or equivalent in other currencies) in monthly revenue, you…

It's silly, but in the LLM world - "open source" is usually used to mean "weights are published". This is not to be confused with the software licensing meaning of "open source".

The more tasteful corners of the LLM world use "open weights" instead of "open source" for licenses that aren't OSI.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#119

Post by pxc »

johnb231 wrote: post #108
irthomasthomas wrote: post #72

Earlier quoted context omitted.

Someone at openai did say it was too big to host at home, so you could be right. They will probably be benchmaxxing, right now, searching for a few evals they can beat.

These are all "too big to host at home". I don't think that is the issue here. https://github.com/MoonshotAI/Kimi-K2/blob/main/docs/deploy_... "The smallest deployment unit for Kimi-K2 FP8 weights with 128k seqlen on mainstream H200 or H20 platform is a cluster with 16 GPUs with either Tensor Parallel (TP) or "data parallel + expert parallel" (DP+EP)." 16 GPUs costing ~$30k each. No one is running a ~$500k server at…

I think what GP means is that because the (hopefully) pending OpenAI release is also "too big to run at home", these two models may be close enough in size that they seem more directly comparable, meaning that it's even more important for OpenAI to outperform Kimi K2 on some key benchmarks.

Re: Kimi K2 is a state-of-the-art mixture-of-experts (MoE) language model

#120

Post by tuananh »

summarity wrote: post #64
vessenes wrote: post #62

I tried Kimi on a few coding problems that Claude was spinning on. It’s good. It’s huge, way too big to be a “local” model — I think you need something like 16 H200s to run it - but it has a slightly different vibe than some of the other models. I liked it. It would definitely be useful in ensemble use cases at the very least.

Reasonable speeds are possible with 4bit quants on 2 512GB Mac Studios (MLX TB4 Ring - see https://x.com/awnihannun/status/1943723599971443134 ) or even a single socket Epyc system with >1TB of RAM (about the same real world memory throughput as the M Ultra). So $20k-ish to play with it. For real-world speeds though yeah, you'd need serious hardware. This is more of a "deploy your own stamp" model, less a "local" mod…

looks very much usable for local usage.
Post reply on HN