Karl Luo Page

PCG C++ Committee Member

• Participate in Tencent engineer promotion assessment; Participate in planning Tencent’s technical blueprint in AI/ML; Contribute code to the Tencent company level infrastructure.

Deep Learning Framework Senior Expert

• Founder and main contributor of open source LLM inference engine Ksana(一念): https://github.com/pcg-mlp/KsanaLLM.

• Member of Tencent’s Hunyuan(混元) LLM as Inference engine developer for LLM model with NVIDIA GPU and Huawei Ascend NPU.

• One of the core founders heterogeneous inference engine, which focuses on LLM and Stable Diffusion inference accelerate for Tencent Zenvideo(智影), QQ AI Agents(QQ AI智能体) and game products.

• One of the core founders of Tencent PCG’s heterogeneous machine learning platform, Numerous (无量), which focuses on large-scale recommendation model training and inference for QQ, Tencent Video, and QQ Music. Numerous (无量) handle over 10 billion queries per day.

• Main Architect of international federated learning framework for GameLoop.

• As HugeCTR contributor, developed in cooperation with the NVIDIA team, broke the world record in performance of the 2020 and 2021 MLPerf training framework.

• Award: 2021, 2022 Excellent Contributor in Tencent Open Source Collaboration Project.

• Award: 2020, 2022, 2023 Tencent Outstanding Contributor.

• Award: 2020 Tencent Excellence in R&D Award.

• Award: 2020, 2021, 2022, 2023 Tencent Excellent Employee.

• Award: 2020 Tencent Open Source Collaboration Award.

Deep Learning Platform Engineer

• Member of Autopilot project

• Developed AI and deep learning platform with: Graph optimization, Pruning, Deep Compression, Quantization, Distillation.

• Tuning operators on TVM with Machine Learning Algorithm like boosting, reinforcement learning.

• Developed TOPI of NPU.

• Research on hardware aware NAS on AutoML-keras.

Deep Learning Software Engineer

• Developed part of feature TensorRT (release: 4.0+, RC: 5.0+).

• cuDNN performance tuning.

• Tuning TensorRT performance with different models.

• Weights Quantification and calibration (INT8, INT4).

• Work in TensorRT project on DrivePX2, Jetson TX2/TX1 platform with Python and C++.

• Work in cuDNN project on DrivePX2, Jetson TX2/TX1 platform with Python and C++.

• Deploy models from other frameworks on TensorRT.

• Optimize inference speed on GPU.

• Develop fast kernel and extra layer plugin.