Skip to content
View Zantares's full-sized avatar
15 followers · 0 following

Achievements

Achievement: QuickdrawAchievement: Pair Extraordinairex2Achievement: YOLOAchievement: Pull Sharkx3Achievement: Arctic Code Vault Contributor

Achievements

Achievement: QuickdrawAchievement: Pair Extraordinairex2Achievement: YOLOAchievement: Pull Sharkx3Achievement: Arctic Code Vault Contributor

Block or report Zantares

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Zantares/README.md

Hi, I'm Lu Teng 👋

Senior AI Compiler Engineer @ EVAS · Hangzhou, China

12+ years across chip compilers, JVM, deep learning frameworks, and AI compilers — from GCC at Huawei to OpenXLA at Intel, now building NPU software stacks from scratch.

🔭 Currently Building: libepu

End-to-end AI compiler stack for a custom NPU via XLA PJRT plugin:

PyTorch / Torch-XLA → StableHLO / HLO → MLIR → VISA → NPU

🚀 Projects I Led

Intel Extension for OpenXLA · Architect · 2023 – 2024

Custom AI compiler on the OpenXLA stack for Intel GPU integration into the upstream community.

  • Designed compiler pipeline: StableHLO → HLO → MLIR/LLVM → SPIR-V
  • Led team to deliver end-to-end GPU support; integrated TF / JAX / PyTorch frontends for LLM & Stable Diffusion workloads
  • Intel GPU liaison to Google OpenXLA community; drove Intel GPU in-tree integration RFC

Intel Extension for TensorFlow (ITEX) · Architect · 2020 – 2023

Custom deep learning framework built on TensorFlow's PluggableDevice / plugin architecture.

  • Led team to develop unified CPU/GPU graph optimization pipeline on top of TF plugin mechanism
  • Frontend op replacement, in-place optimization, dynamic Eigen/oneDNN runtime tuning

🌱 Representative Upstream Contributions

Selected PRs and RFCs with architectural impact or substantial code changes — not exhaustive.

Project Contribution
OpenXLA Community RFC: Intel GPU in-tree integration — proposed Intel GPU as first-class OpenXLA device; 40+ community review threads
PyTorch/XLA Composite op lowering for training — lower Aten ops to composite kernels instead of fine-grained ops for end-to-end training performance
TensorFlow BF16/FP32 RandomUniform SIMD vectorization — Eigen packet vectorization for low-precision random ops; core enabler for BF16 inference path

View all pull requests →

💼 Experience

2024 – now   EVAS          Senior AI Compiler Engineer · libepu tech lead
2018 – 2024  Intel         AI Framework Engineer / Architect · OpenXLA, ITEX, TF CPU
2012 – 2018  Huawei        Compiler Engineer · GCC chip compiler, JVM, Ascend ABI

🛠 Tech Stack

C/C++ · MLIR/LLVM · XLA / OpenXLA / StableHLO · PyTorch / Torch-XLA / PJRT · TensorFlow / JAX · GCC · Bazel

📫 Contact

Pinned Loading

  1. skills skills Public

    My AI tool skills.

  2. torch-xla torch-xla Public

    Forked from pytorch/xla

    Enabling PyTorch on XLA Devices (e.g. Google TPU)

    C++

  3. pytorch/xla pytorch/xla Public

    Enabling PyTorch on XLA Devices (e.g. Google TPU)

    C++ 2.8k 573

  4. bert bert Public

    Forked from google-research/bert

    TensorFlow code and pre-trained models for BERT

    Python 1