alt Hacker NewsDeepInfer: High-performance inference framework for DeepSeek models
Comments
If that's the thing that makes $0.00X cached inputs possible, then this is huge!
The quality of technical discussions here has clearly nosedived over the past few years. This release basically confirms it.
Open source is different from open weights
Well played!
I don't understand what is in it for them. I understand that the government wants openness but who's paying them to give away their secrets?
DeepSeek doesn't really care about models, they're aiming to become a platform: https://try.works/a-follow-up-on-china's-open-ai-race
They open source it because they don't need it anymore. Probably they are doing something better with Huawei
the only advantage i see long run is they can get free contributions world wide to lower their engineering costs. In the short run, frontier labs gets to use some of the ideas.
Need to see a Strata style take on this.
hi
Just another tour de force from DeepSeek.
I'm not saying that the US labs (Anthropic, OpenAI, and Meta) aren't pulling their weight, they are all doing amazing work as well, but Deepseek may belong in that tier.
So what? We already have multiple open source inference engines for DeepSeek models - that's how there are over 30 providers already running DeepSeek around the world. There's no secret sauce here, they've been publishing everything they do in papers from day 1.
[flagged]
[flagged]
[dead]
[flagged]
[flagged]
[dead]
[dead]
Deepseek released just enough to stir the pot but not enough to cook your own from scratch.
Wake me up when we can clone, point it at a dozen H100s we own, and get a working production service.
Anthropic and OpenAI will do just fine.
they must know something is coming from the american labs for some time, you don't release this just like that!