Moore Threads First to Complete Day-0 Adaptation of the Step 5 Preview Large Model from StepFun

Moore Threads First to Complete Day-0 Adaptation of the Step 5 Preview Large Model from StepFun

AIbase Base

The ecosystem integration of large models and domestic hardware is pressing the fast-forward button. On October 15, StepFun officially open-sourced its new-generation large model Step5Preview. On the same day, Moore Threads officially announced that its MTT S5000 AI computing card and the accompanying MUSA software stack had completed Day-0 adaptation of the model at the first opportunity, demonstrating extremely high responsiveness and iteration efficiency.

In terms of technical specifications, Step5Preview adopts an advanced sparse MoE architecture, with a total parameter count of 600B and 27B parameters activated per token, and natively supports an ultra-long context window of up to 1 million tokens. Backed by powerful underlying hardware and software support, Moore Threads' adaptation work this time deeply covered core aspects such as hybrid attention, mixed-precision inference, and speculative decoding.

This efficient collaboration between the two parties not only marks a new milestone in the ability of domestic AI computing hardware to support the cutting-edge ultra-large model ecosystem, but also provides developers with a mature and reliable underlying deployment path for deploying and running complex long-text large models on domestic computing platforms.

This article comes from the AIbase Daily

Welcome to the [AI Daily] column! Here is your daily guide to exploring the world of artificial intelligence. Every day we present you with hot topics in the AI field, focusing on developers and helping you gain insight into technology trends and learn about innovative AI products and applications.