Nov 13 | Speeding Up Training with FP8 and Triton | ML OpenTalk

[Tatev Vardanyan]

Dear community, this week we will skip Friday reading group and instead please join us on Nov 13th for and MLOpen Talk at YandexHall. Details are below.

Join us for the next ML OpenTalk event, co-organized and hosted by Yandex Armenia on November 13. Vladislav Savinov, Team Lead at YandexGPT pretraining, will share how modern research labs push the limits of GPU efficiency with FP8 precision and Triton. 

Topic: Speeding Up Training with FP8 and Triton

Date & Time: November 13, 7PM

Location: Yandex Hall, 35 Moskovyan Street

Registration Link: https://forms.yandex.ru/surveys/13794306.bdd0707f942d9c6592938b40d89b1b6a5d672933/

In this session, you'll discover:

• Why training in lower precision can cut costs and speed up compute.
• What recent FP8 research papers and open-source tools reveal.
• How GPUs actually work and how Triton kernels boost performance in practice.
• Real insights from YandexGPT’s production-scale FP8 pretraining.

We are looking forward to seeing you on November 13th. 

Regards,

Tatev