HY

Personal site

Hou Yuteng侯宇腾

High-performance kernels on ARM SME/SVE, and AI systems that actually run. Graduated from Harbin Institute of Technology, Shenzhen, and KU Leuven (Cum Laude).

View projects Email me

01

About

GEMM and attention on ARM SME/SVE, and HPC, AI agents, and model deployment brought together into systems that run.

  • 01 ARM / HPC

    UniGEMM, GEMM, and attention on SME/SVE

  • 02 AI agents

    Tool use, remote execution, and agent workflows

  • 03 Model deployment

    Local inference, ARM platforms, and performance checks

Education

  • 2019–2024Harbin Institute of Technology, Shenzhen · B.Eng. in Electronics and Information Engineering
  • 2021–2023KU Leuven · B.Sc. in Engineering Technology, joint program
  • 2023–2024KU Leuven · M.Sc. in Engineering Technology, Software Systems · Cum Laude

Experience

  • 2025–presentUniGEMM, GEMM, and attention on ARM SME/SVE: multi-precision assembly generation, core GEMM kernels above 95% of theoretical peak, and memory-access and compute optimizations across several attention variants.
  • Same periodMPI and UCX performance work in HPC settings, and maintenance of KBLAS and Eigen.
  • 2023–2024Stability after model deployment: monitoring input distributions and output anomalies to reduce concept drift.

02

Projects

A few public projects

View all repositories
01

cannbot-gateway-proxy

A local API in front of the CANNBot gateway. OpenAI requests pass through unchanged; Anthropic Messages are translated in both directions. Credentials are injected at call time, so a local agent does not keep secrets in its config. An admin console shows usage and latency such as TTFT and TPOT.

JavaScript 02

cpu_model_deploy

Offline deployment of Qwen3.6-35B-A3B on the Kunpeng 920 (ARM CPU). Built and run with llama.cpp, with no Python runtime.

C++ 03

agent_remote

A remote-execution framework for AI-assisted development. The build machine submits tasks and artifacts; a worker on the execution machine pulls them, runs them, and returns logs. When the two sides cannot reach each other, a relay queue carries the handoff.

Python 04

gemmtest

Numerical checks for GEMM on ARM. One codebase covers macOS development machines and ARM64 servers, with a compile-time switch for HBM and comparisons across FP32, FP16, and BF16.

C++ 05

LLM_papers

Chinese reading notes on large-model papers, arranged by topic — from surveys through architecture and training — with explanations of the harder ideas.

Notes 06

agent_knowledge

A library of Chinese technical notes. Each topic is layered as a design note, an outline, and a finished draft. The current subject is low-level CPU software in the agent era, with figures cited to their sources.

HTML

03

Contact

Say hello

Have an idea, or want to talk?
Send a note.

GitHub · @0tomhoyt