AI People Wiki

Builders & open source

Aaron Teo

Technology Engineer, IBM Client Engineering, IBM

  • Technology Engineer, IBM Client Engineering, IBM1
  • Code owner for the zDNN backend; member of the CI team, llama.cpp project2

Aaron Teo is a Technology Engineer in IBM Client Engineering whose GitHub profile lists AI on IBM Z, inference engines, and IBM Z and LinuxONE as areas of work. llama.cpp's CODEOWNERS file lists Teo, as the ggml-zdnn team, among the owners of the zDNN backend, which enables llama.cpp on IBM z17 and later systems through IBM's zDNN acceleration library, and as a member of the CI team. Teo last updated the project's build guide for IBM Z and LinuxONE in September 2026, and Teo's commits add s390x CPU optimizations, zDNN operations and CI coverage. Teo also publishes a Hugging Face collection of big-endian GGUF models, including IBM Granite models, verified to run on IBM z15 and later systems.123456

At a glance

Current roles
Technology Engineer, IBM Client Engineering +1 more
Sector
Company
Work listed
2
Sources cited
10
Profile reviewed

Open-source work

Aaron Teo's work on llama.cpp brings the inference engine to IBM Z and LinuxONE mainframes. Teo's first merged pull request, in February 2025, added support for the s390x SIMD instruction set to the ggml CPU backend, followed by changes such as big-endian model conversion and z17 compile detection.785

In August 2025 Teo's pull request introducing an initial IBM zDNN backend was merged. The backend uses IBM's zDNN library to run inference on the Neural Network Processor Assist accelerator in IBM Telum processors, and llama.cpp's CODEOWNERS file lists Teo, as the ggml-zdnn team, among its owners and as a member of the CI team.932

Teo's recent commits, through October 2026, add operations to the zDNN backend such as flash attention and Q8_0 matrix multiplication, vectorize CPU activation functions, and update the zDNN and s390x build documentation. Teo also maintains a Hugging Face collection of big-endian GGUF models verified to run on IBM z15 and later systems.546

Work and contributions

  • llama.cpp support for IBM Z and LinuxONE Open-source code

    CODEOWNERS lists Teo as the ggml-org/ggml-zdnn team, which co-owns ggml/src/ggml-zdnn, and in ggml-org/ci. The s390x build guide credits Teo with its September 2026 update. Commits authored from July to October 2026 add Q8_0 matrix multiplication and memory fixes to the zDNN backend, s390x vector and repack optimizations, CI coverage, and Granite parameter-count fixes. The repository is MIT-licensed.24510

  • s390x Verified Models (Hugging Face collection) Open weights

    A collection owned by Teo's Hugging Face account of big-endian GGUF models, including IBM Granite and Qwen2.5 models, whose tokenizers were verified to run on IBM z15 and later systems; the s390x build guide points to it for pre-converted models.64

Timeline

  1. Added s390x SIMD instruction set support to the llama.cpp CPU backend.78
  2. Added z17 compile detection to the ggml CPU backend.8
  3. Merged the initial IBM zDNN backend into llama.cpp.98
  4. Added a flash attention operation to the llama.cpp zDNN backend.5

Where they work

Suggest a correction or removal

Profiles cover public professional work only, and each statement cites a source. If something is out of date, missing or wrong, tell us and link a source. People can also ask for their profile to be corrected or removed.

References

10 sources · profile reviewed Oct 21, 2026 · data updated Oct 21, 2026

  1. taronaeo (Aaron Teo) on GitHubGitHub · Official page · accessed Oct 21, 2026
  2. ggml-org/llama.cpp CODEOWNERSggml-org (GitHub repository) · Repository · accessed Oct 21, 2026
  3. llama.cpp for IBM zDNN Accelerator (docs/backend/zDNN.md)ggml-org (GitHub repository) · Documentation · accessed Oct 21, 2026
  4. llama.cpp build guide for IBM Z and LinuxONE (docs/build-s390x.md)ggml-org (GitHub repository) · Documentation · published Sep 8, 2026 · accessed Oct 21, 2026
  5. ggml-org/llama.cpp commits authored by taronaeoggml-org (GitHub repository) · Repository · accessed Oct 21, 2026
  6. s390x Verified Models — Hugging Face collection by taronaeoHugging Face · Official page · accessed Oct 21, 2026
  7. ggml-cpu: Support s390x SIMD Instruction Set (ggml-org/llama.cpp pull request #12019)ggml-org/llama.cpp (GitHub repository) · Repository · published Feb 22, 2025 · accessed Oct 21, 2026
  8. ggml-org/llama.cpp merged pull requests authored by taronaeoggml-org/llama.cpp (GitHub repository) · Repository · accessed Oct 21, 2026
  9. ggml: initial IBM zDNN backend (ggml-org/llama.cpp pull request #14975)ggml-org/llama.cpp (GitHub repository) · Repository · published Aug 15, 2025 · accessed Oct 21, 2026
  10. ggml-org/llama.cpp LICENSE (MIT)ggml-org (GitHub repository) · License · accessed Oct 21, 2026