Vladimir Zeljkovic retweeted
Check out our work with @OrukLabs and @openwhispr
Model that beats @nvidia deserves to run fast on @Apple .
We collaborated with @OrukLabs and @openwhispr to bring the best voice model to your local device.
@oruklabs has built the best transcription model. @openwhispr has built the best engine for running it. @hoid_ai has made it fast.
@nrol_ling @gabrielste1n @ppadjin @Zelja__
171 ms to 110 ms.
That's Orukeet, @OrukLabs's new open speech model, after going through @hoid_ai.
It beats @nvidia's Parakeet, covers 25 languages, and is now the default transcription model in @openwhispr (100k+ downloads)
step aside @nvidia, there's a new top dog in speech models: orukeet.
trained by @OrukLabs, optimized by @hoid_ai, and available on day zero in @openwhispr (100k+ downloads).
25 languages. Open weights + code.
Vladimir Zeljkovic retweeted
Orukeet wins 61 of 74 benchmark splits; Parakeet wins 13.
Hoid's latest OpenWhispr PR: 171 → 110 ms for Orukeet (36% lower warm median).
Apple M5, 11-second clip, before/after optimization.
github.com/OpenWhispr/openwh…
oruk.ai/research/orukeet
Vladimir Zeljkovic retweeted
Hoid is discovering megakernels that are too fast too handle. More results and a blog soon.
Let's make that model faster!
it's been a pleasure working with @emilybenn12, @custo_lejla , @_CallMeMacy, @andrewchen, @Chen, @justmazer, @tmhammer, @nazzari, the @speedrun team!!
If this sounds interesting to you, our DMs are open
tl;dr of life recently: @ppadjin , @Zelja__ and I quit our jobs to build a startup (link in comments) and we're backed by @speedrun . It is the best time in history to build a company and we didn't plan on missing the boat.
@hoid_ai is an agentic AI compiler that turns any AI model into a highly optimized implementation for the given hardware.
We’re building the layer between AI models and the silicon they run on, using AI itself to explore the optimization space and squeeze out every last bit of performance.
If you're in the AI inference space hit me up let's chat.
