Mengni Wang
Software Engineer at Intel
Chandler, Arizona, United States
Summary
🤩
RockstarMengni Wang is a software engineer with five years of experience specializing in ML model optimization and deployment, currently contributing at Intel from Chandler, Arizona. She has hands-on expertise in quantization and sparsity techniques for LLMs and vision models, having added and integrated numerous int8 models (ResNet50, VGG16, BERT, etc.) into the ONNX models collection. Her contributions to Intel’s neural-compressor project improved unit testing, configuration templates, and multi-format inference support, reflecting a strong focus on making state-of-the-art compression techniques production-ready. Comfortable working across TensorFlow, PyTorch, and ONNX Runtime, she bridges research and engineering to boost inference performance and reduce model footprint. Colleagues rely on her pragmatic problem-solving and attention to accuracy when enabling low-bit deployments at scale.
5 years of coding experience
Github Skills (15)
quantization10
sparse10
machine-learning10
eval10
deep-learning10
onnx10
large-language-models10
evaluation10
pre-trained-model10
model-optimization10
computer-vision9
tensorflow9
pytorch9
unit-testing8
nlp7
Programming languages (3)
Github contributions (5)
intel/neural-compressor
Jul 2020 - Jan 2023
SOTA low-bit LLM quantization (INT8/FP8/INT4/FP4/NF4) & sparsity; leading model compression techniques on TensorFlow, PyTorch, and ONNX Runtime
Role in this project:
ML Engineer
ML EngineerContributions:118 reviews, 246 commits, 79 PRs in 2 years 6 months
Contributions summary:Mengni primarily worked on enhancing the `intel/neural-compressor` repository, focusing on improving the performance and functionality of the model, particularly concerning quantization and sparsity techniques for large language models (LLMs). Their contributions included adding unit tests, fixing configuration templates, modifying inference and benchmark files, enabling features for model formats (e.g., ckpt), and addressing various issues related to metrics and model accuracy. These efforts were centered around optimizing and deploying quantization-aware models, making them more efficient.
knowledge-distillationauto-tuningcompressorsparsityintel
onnx/models
Jul 2021 - Dec 2022
A collection of pre-trained, state-of-the-art models in the ONNX format
Role in this project:ML Engineer
ML EngineerContributions:5 reviews, 16 commits, 15 PRs in 1 year 4 months
Contributions summary:Mengni primarily contributed to the integration and addition of pre-trained, quantized (int8) models within the ONNX models repository. Their commits include the addition of various int8 models like ResNet50, VGG16, ShuffleNetV2, BERT, and others, demonstrating a focus on model optimization and deployment. The user also updated existing model configurations, README files, and test data to accommodate the newly added and updated quantized models. This work helps to improve the inference performance of these models.
downloadpytorchartdeep-learningpre-trained
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.