Romain Lepert
@obust.bsky.social
deep learning, programming languages, synthesizers, sound...
Reposted by Romain Lepert
We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1
And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder
[1/3]
And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder
[1/3]
March 12, 2025 at 1:22 PM
We've kept pushing our Open-R1 project, an open initiative to replicate and extend the techniques behind DeepSeek-R1
And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder
[1/3]
And even we were mind-blown by the results we got with this latest model we're releasing: ⚡️OlympicCoder
[1/3]
This phone made my night in the forest back in 2019
my dear friend and artist Boyan (visual artist for the "ethereal spaces" video on my YouTube channel) compiled the interactive visual works he created for the parties we co-organize in the forest by amsterdam. enjoy the trippy and funny worlds imagined by TheKnowDough. youtu.be/OS7AvoilfR8
Boyan Mihaylov / TheKnowDough - interactive visual installations for Activités en Forêt 2019 - 2024
YouTube video by WP Frequencies
youtu.be
December 21, 2024 at 6:25 PM
This phone made my night in the forest back in 2019
Proud to present our work on optimizing Mixture of Experts models for on-device generation speed: arxiv.org/pdf/2412.00099
We introduce a cache-aware routing that reduces cache misses of MoEs, improving token generation throughput by 2×. Perfect for memory-constrained devices.
We introduce a cache-aware routing that reduces cache misses of MoEs, improving token generation throughput by 2×. Perfect for memory-constrained devices.
arxiv.org
December 4, 2024 at 4:02 PM
Proud to present our work on optimizing Mixture of Experts models for on-device generation speed: arxiv.org/pdf/2412.00099
We introduce a cache-aware routing that reduces cache misses of MoEs, improving token generation throughput by 2×. Perfect for memory-constrained devices.
We introduce a cache-aware routing that reduces cache misses of MoEs, improving token generation throughput by 2×. Perfect for memory-constrained devices.