原帖引用 · 原文
Rohan Paul@rohanpaul_ai
@hassanience Notable that they tested running all-reduce on a separate stream and dropped it because it competed for memory bandwidth. Useful negative result.
not long ago, people questioned whether AMD GPUs could keep up. today, we’re setting records on them. 1,129 output tps at P90. Real production traffic, not a toy project.
a little faith, a lot of engineering. here’s how we did it. https://t.co/3boDmk3LQF
在 X 查看这条帖子 ↗
· UTC+8