Cheng Yao

Cheng Yao works at AMD’s Training at Scale team, mainly focus on optimizing training performance and developing distributed systems for large language models.

Posts by Cheng Yao

October 21, 2026

MoE Training Optimization with Primus

Learn how to speed up MoE training on AMD Instinct GPUs with Primus: a fused MoE megakernel, FP8/MXFP8 expert GEMMs, and training-at-scale tuning.

https://rocm.blogs.amd.com/software-tools-optimization/primus-moe-optimization/README.html

September 03, 2026

Enabling DeepSeek-V4-Flash Training on AMD Instinct MI355X GPUs with Primus

DeepSeek-V4-Flash training on AMD Instinct GPUs with Primus: model architecture introduction, performance projection, kernel optimizations, and how to reproduce.

https://rocm.blogs.amd.com/software-tools-optimization/primus-deepseek-v4/README.html

July 06, 2026

Primus Tuning Agent: Closing the Configuration-Search Loop

Use the Primus Tuning Agent to automatically find optimal LLM training configurations on AMD Instinct GPUs.

https://rocm.blogs.amd.com/software-tools-optimization/primus-tuning-agent/README.html

April 24, 2026

Primus Projection: Estimate Memory and Performance Before You Train

Learn how to use the Primus projection tool to estimate memory and performance for large-scale LLM training on AMD Instinct™ accelerator platforms.

https://rocm.blogs.amd.com/software-tools-optimization/primus-projection/README.html

February 23, 2026

Primus-Pipeline: A More Flexible and Scalable Pipeline Parallelism Implementation

Learn how to use our flexible and scalable pipeline parallelism framework with Primus backend and AMD hardware.

https://rocm.blogs.amd.com/software-tools-optimization/primus-pipeline/README.html

December 16, 2025

MoE Training Best Practices on AMD GPUs

Learn how to optimize Mixture-of-Experts (MoE) model training on AMD Instinct GPUs with ROCm. Maximize your AI training performance now!

https://rocm.blogs.amd.com/software-tools-optimization/primus-moe-package/README.html