Daniël De Kok
Machine Learning Ops Engineer at Hugging Face
Groningen, Groningen, Netherlands
Summary
🤩
Rockstar🎓
Top SchoolDaniël De Kok is a Machine Learning Ops Engineer with 16 years of experience blending deep NLP research and systems-level programming to deliver high-performance ML and inference solutions. He has driven performance and numerical robustness across major open-source projects—most notably improving Thinc and integrating flash-attention GPT-2 in Hugging Face’s text-generation-inference—while working in Python, Rust, C/C++ and CUDA. At Explosion he led performance work for spaCy and Thinc, added Apple MPS support, and designed Curated Transformers used for spaCy’s transformer pipelines. Comfortable from low-level memory management to mixed-precision GPU kernels, he focuses on turning research-grade models into efficient production systems. Based in Groningen, he pairs a PhD in Computational Linguistics with sustained OSS contributions (including TensorFlow and Rust bindings), and outside work he’s an avid cyclist.
16 years of coding experience
16 years of employment as a software developer
Doctor of Philosophy - PhD, Computational Linguistics, Graduated with distinction, Doctor of Philosophy - PhD, Computational Linguistics, Graduated with distinction at University of Groningen
Github Skills (43)
transformers10
pytorch10
performance-analytics10
performance-monitor10
parser10
python10
operation10
multithreading10
tensorrt10
memory-management10
machine-learning10
performance-measurement10
numpy10
binding10
flash10
Programming languages (18)
Github contributions (5)
huggingface/text-generation-inference
May 2024 - Mar 2025
Large Language Model Text Generation Inference
Role in this project:
Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:254 reviews, 265 PRs, 531 pushes in 10 months
Contributions summary:Daniël primarily contributed to the implementation of GPT-2 models with flash attention, adding the necessary code to support and integrate these models. They focused on the integration of the flash attention mechanism, modifying existing code and introducing new files for the FlashGPT2ForCausalLM model. The user also fixed issues related to GPTQ models when non-float16 data types are used and added a test to ensure the integration is working correctly. The user has demonstrated a good grasp of machine learning model integration.
nlppytorchlanguage-modelbloombert
explosion/thinc
Aug 2021 - Jan 2023
🔮 A refreshing functional take on deep learning, compatible with your favorite libraries
Role in this project:Back-end Developer & ML Engineer
Back-end Developer & ML EngineerContributions:8 releases, 310 reviews, 124 commits in 1 year 5 months
Contributions summary:Daniël primarily focused on improving the `thinc` library by addressing critical issues related to numerical computation and performance optimization within the framework. Their contributions included fixing bugs in the `numpy_ops gemm` and enhancing the layer functionality to support GPU profiling using NVTX ranges. Furthermore, the user implemented features to support mixed-precision training in the PyTorch shim, and added support to handle sequence lengths in `seq2col`. These efforts resulted in significant speed improvements and enhanced flexibility in core numerical operations.
pythonfunctional-programmingtensorflowtype-checkingnatural-language-processing
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.