Tyler Smith
Member Of Technical Staff at vLLM
Cambridge, Massachusetts, United States
Summary
🤩
Rockstar🎓
Top SchoolTyler Smith is a Member of Technical Staff at Red Hat and a vLLM committer with 12 years of experience building high-performance ML and numerical software. He holds a PhD in Computer Science from UT Austin and has led engineering teams at Neural Magic, progressing from software engineer to technical director. Tyler’s hands-on work focuses on inference engineering and GPU-optimized kernels—contributing CUTLASS FP8 and sparse GEMM integrations to the widely used vLLM project and performance fixes in the BLIS framework for Xeon Phi. He blends deep research experience from ETH Zurich and academia with production-grade system design, shipping optimizations that improve both throughput and memory efficiency. Known for tackling low-level kernel refactors and CUDA graph compatibility issues, he brings rare expertise at the intersection of ML inference, compilers, and high-performance computing. Based in Cambridge, MA, he combines leadership, open-source stewardship, and a proven track record of turning advanced numerical ideas into deployable inference engines.
12 years of coding experience
12 years of employment as a software developer
Doctor of Philosophy (Ph.D.), Computer Science, Doctor of Philosophy (Ph.D.), Computer Science at The University of Texas at AustinBachelor’s Degree, Computer Science, Bachelor’s Degree, Computer Science at Purdue University
English
Github Skills (15)
c1710
cuda10
matrix-multiplication10
pytorch10
p810
hpc10
c1110
blas10
linear-algebra10
optimization10
multithreading9
microkernel9
openmp9
deep-learning9
machine-learning8
Programming languages (6)
Github contributions (5)
vllm-project/vllm
Mar 2024 - Apr 2025
A high-throughput and memory-efficient inference and serving engine for LLMs
Role in this project:
ML Engineer
ML EngineerContributions:853 reviews, 175 PRs, 96 pushes in 1 year
Contributions summary:Tyler's commits primarily focus on integrating and optimizing CUTLASS kernels within the vLLM project, specifically for FP8 and sparse matrix multiplications. Their contributions involve implementing and refactoring kernels to support various quantization schemes, including block-wise and per-token quantization, as well as optimizing performance for different GPU architectures. The user also addressed CUDA graph compatibility and fixed issues related to the integration of these kernels into the model execution pipeline.
amdcudagpthpuinference
flame/blis
Feb 2014 - Jun 2018
BLAS-like Library Instantiation Software Framework
Role in this project:Backend Developer
Backend DeveloperContributions:86 commits, 6 PRs, 13 pushes in 4 years 4 months
Contributions summary:Tyler contributed to the BLIS library, a framework for high-performance BLAS-like libraries, specifically by optimizing and fixing bugs in the microkernel implementations for the Xeon Phi architecture. Their work focused on improving the performance of the gemm (general matrix-matrix multiplication) operation, including code modifications, optimization of prefetch instructions, and correcting issues related to scattered updates. The user also added multithreading infrastructure and incorporated performance optimizations within the packm (packing of matrices) and trmm/trsm operations.
software-frameworklinear-algebra-librarylinear-algebramatrix-multiplicationblis
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.