Haishuo Kong

Haishuo Kong is an MTS Software Development Engineer on the Training at Scale team, focused on building observability systems for large-scale training clusters. He has developed observability platforms for both CPU and GPU environments and has extensive experience supporting large-scale model training workloads.

Posts by Haishuo Kong

Resilient Large-Scale Training: Integrating TorchFT with TorchTitan on AMD GPUs

Achieve resilient, checkpoint-less distributed training on AMD GPUs by integrating TorchFT with TorchTitan on Primus-SaFE.

February 08, 2026 by Chaojun Hou, Lei Zhang, Lei Wei, Xiaofeng Zheng, Haishuo Kong, Jingxian Wang, Wen Xie, Liz Li, Yao Fu, Zhenyu Gu