Baidu’s Baige team successfully adapted the GLM‑4.x series language models to the Kunlun XPU platform by leveraging SGLang and the vLLM‑Kunlun plugin, employing agile adaptation, precision alignment with torch_xray, and extensive performance tuning to achieve GPU‑level accuracy and superior inference speed.
In just two days, Baidu Baige and Kunlun's engineers extended the vLLM‑Kunlun Plugin to overcome asymmetric KV dimensions and integrate SWA+Sink attention, achieving lossless, high‑performance inference of the MiMo Flash V2 model on the Kunlun P800 XPU.
The vLLM‑Kunlun Plugin, built on the vLLM hardware‑plugin RFC, lets developers deploy any major large language model on Baidu's Kunlun XPU instantly without modifying vLLM core code, dramatically shortening migration time, providing high‑performance fusion operators, and offering open‑source tools for precision verification and profiling.
This article examines Broadcom's Scale Up Ethernet (SUE) framework, detailing how it addresses AI/HPC rack‑scale interconnect challenges by delivering ultra‑high bandwidth, microsecond‑level latency, memory‑semantic operations, and seamless compatibility with existing Ethernet infrastructure for large XPU clusters.
As AI models grow to massive scales, the demand for ultra‑high‑performance, low‑latency networking in xPU clusters intensifies, prompting a shift from dense to MoE architectures and driving the evolution of Scale‑up networks, where Alibaba Cloud’s UPN design tackles bandwidth, cost, and reliability challenges.
This article explains the concepts of AI Pods and GPU clusters, compares vertical (scale‑up) and horizontal (scale‑out) expansion, describes XPU types, discusses internal and inter‑pod communication, and evaluates the benefits and drawbacks of each scaling approach along with relevant networking technologies.
Kunlun Chip’s XPU‑R architecture combines high‑performance SDNN and Cluster compute units, 512 GB/s GDDR6 memory, and PCIe 4.0 interconnect, supported by an LLVM‑based software stack, CUDA‑like programming model, and seamless PaddlePaddle integration, enabling efficient AI training and inference with significant cost and performance gains.
