Senior AI Graph Compiler Engineer

Modular•United States / Canada,

21h•$198,000 - $286,000•Remote

About The Position

The graph compiler is a central piece of MAX, which executes ML models at state-of-the-art performance across multiple hardware platforms. It is built on MLIR from the ground up using a novel design based on tight integration with the Mojo programming language. Modular’s compiler stack is vertically integrated, from ML serving layer, through the Python API, down to highly tuned kernel implementations - opening up opportunities for optimization not available in competing technologies. This role involves extending and developing the graph compiler to support state-of-the-art features and hardware platforms. You will improve the performance and usability of the graph compiler, creating a joyful developer experience. Modular’s vertically integrated ML serving stack allows you to solve a uniquely diverse set of problems, like heterogeneous compilation, kernel fusions, and distributed model execution.

Requirements

5+ years of software engineering experience.
Proficiency in C++.
Experience working with compilers for machine learning frameworks.
Creativity and curiosity for solving complex problems, a team-oriented attitude that enables you to work well with others.

Nice To Haves

Knowledge of and experience working with MLIR and LLVM.
Experience with ML parallel / distributed programming, heterogeneous ML computation, and/or code generation.
Knowledge of basic ML implementation and modeling techniques and familiarity with ML frameworks like PyTorch, JAX, or TensorFlow.

Responsibilities

Design and develop compiler optimizations to improve inference efficiency across CPUs, GPUs, and ML accelerators, including kernel fusion, memory planning, and shape optimization.
Compile large-scale application workloads onto heterogeneous hardware: map hardware-agnostic graphs of computational operators onto hardware-specific graphs of devices and nodes.
Collaborate with the kernels, serving, and Mojo compiler teams to implement core technologies that deliver end-to-end performance across a variety of heterogeneous hardware platforms.
Write design documents for new features and drive for alignment on designs with cross-functional teams.
Collaborate with the customer success team and engage with customers to understand their performance requirements and use cases.