Principal MTIA Compiler Architect — AI Accelerator Stack

Meta; New York (NY); Vor Ort; USD 219.000 - 301.000; Vollzeit

Zusammenfassung

Meta is seeking a principal-level Compiler Architect to work across the full compiler stack for MTIA (Meta Training and Inference Accelerator). You will design and implement compiler infrastructure that enables efficient code generation, optimization, and execution on Meta's custom AI accelerators, spanning graph transformations through intermediate representations to low-level code generation.

This role targets engineers who build the foundational software stack to unlock performance for

Qualifikationen

• Bachelor's degree in Computer Science, Computer Engineering, or equivalent practical experience.
• Experience in compiler development, code generation, or performance optimization for accelerators.
• Experience architecting production compiler infrastructure for ML accelerators, GPUs, or custom silicon.

Aufgaben

• Design compiler architecture and contribute across the full MTIA compiler stack, including graph compilers, IRs, optimizations, and code generation.
• Build extensible compiler frameworks and IRs for rapid iteration amid evolving ML model architectures.
• Identify and eliminate performance bottlenecks across the compiler stack from high-level graph optimizations to code gen.

Kenntnisse

Compiler development

Code generation

Performance optimization

Cross-functional collaboration

Mentoring engineers

Ausbildung

Bachelor's degree in Computer Science/Engineering

Tools

MLIR

Jobbeschreibung

Meta is seeking a principal-level Compiler Architect to work across the full compiler stack for MTIA (Meta Training and Inference Accelerator). You will design and implement compiler infrastructure that enables efficient code generation, optimization, and execution on Meta's custom AI accelerators, spanning graph transformations through intermediate representations to low-level code generation.

This role targets engineers who build the foundational software stack to unlock performance for