CEO & Co-Founder of ai&. Ex-CCO@Tenstorrent. 元社外取締役@Sanrio. 元レノボ兼NEC PC代表取締役執行役員社長. Ex-VP@Lenovo, CEO@NEC PC, CVP@AMD. Living in Austin🇺🇸なう. Ex-🇨🇦🇸🇬🇯🇵.
先月出した本について、色な方から聞かれてますので、もう一度リンクを載せさせて頂きます😅🙏🏻 因みに、ビジネスの本ではありません。面白い日本語のトリビアの本です。まさに、僕にとって気になる日本語です。Thank you for asking about my new book! https://t.co/pwHjQtA1wK
Proud to share ai&’s partnership with @NEC to help them reduce costs on their AI inference spend. They have started to use tokens from Japan’s largest dedicated inference service. ai& ( @aiand_ ) runs open weight models exclusively in our 🇯🇵 datacenters.
https://t.co/s2EDu7gYPi
Check out what @aiand’s brilliant @Weyaxi cooked up for God’s presentation at SGLang’s (@sgl_project) event in SF this week.
ai& was proud to participate and it was a great event.
Way to go Yagiz ❤️
We wrote up RL on heterogeneous hardware at @aiand_, the work I presented at @sgl_project Summit today: train on one chip, roll out on any. One trainer, rollouts from four chips across two vendors, all over plain HTTP.
A two-node B300 trainer, fed by B300, H200, MI355X and MI300X engines for ten hours. Every adapter sync landed on every engine. Zero restarts.
Thread below 🧵
乾杯 to Sake & Inference!
Japan is a long flight from SF, so we brought a little of it with us to help close out @sgl_project Summit week. Thanks to everyone who joined us for great sake, conversations and new connections.
You bring the AI. We bring Japan.
#AI #Inference #Japan
had so much fun presenting at sglang summit! thanks to everyone who came to the talk and stopped by our booth, and to @sgl_project@radixark@lmsysorg for having us 🙏
What if image recognition could unlock the classics? 📖 A Thai-born researcher captivated by Japanese literature built an AI app that makes Japan’s 1,000-year-old documents accessible to everyone, not just specialists. How? 🔍 Find out more👇
What if image recognition could unlock the classics? 📖 A Thai-born researcher captivated by Japanese literature built an AI app that makes Japan’s 1,000-year-old documents accessible to everyone, not just specialists. How? 🔍 Find out more👇
We ran RL on heterogeneous hardware: one B300 trainer fed by B300, H200, MI355X and MI300X rollout engines for ten hours over plain HTTP, with every adapter sync landing on every engine. Presented at @sgl_project Summit.
Full write-up: https://t.co/DcNE8BhDY3
RL rollouts don't have to run on the same hardware as training.
@Weyaxi and the @aiand_ team showed how Miles can train on B300s, while SGLang serves rollouts across NVIDIA B300/H200 and AMD MI355X/MI300X.
Great work, and thanks for sharing at SGLang Summit!