Software Engineer, MTIA SW Performance Autotuning

Job Title

Software Engineer, MTIA SW Performance Autotuning

Role Summary

Lead the MTIA Software Performance Autotuning team to automate performance tuning for Meta's MTIA training and inference accelerator. Own the autotuning strategy, search and benchmarking infrastructure, and cross-team integrations to maximize runtime performance across compiler, kernel, and runtime layers.

Experience Level

Mid-level — 4+ years experience in ML systems, AI infrastructure, or performance engineering.

Responsibilities

Work as a technical leader to design and deliver autotuning systems that find high-performance configurations automatically and at scale.

• Define and own performance autotuning strategy for MTIA across current and future chip generations.
• Design and implement autotuning infrastructure: search spaces, cost models, benchmarking harnesses, and regression detection.
• Balance search cost versus performance gain and integrate autotuning into compile/deploy pipelines.
• Partner with compiler, kernel, and runtime teams to expose tunable parameters and encode automated searches.
• Debug cross-layer performance issues spanning kernel, runtime, compiler, and serving infrastructure to identify root causes.
• Collaborate with product and ML teams to convert performance headroom into delivered speedups for production models.
• Set technical direction, mentor engineers, and help grow the team's impact.

Requirements

Must-have technical skills and experience to perform immediately in the role. Preferred skills listed separately.

• Strong Python and C++ skills with hands-on experience across the PyTorch stack.
• 4+ years in ML systems, AI infrastructure, performance engineering, or comparable roles.
• Experience driving problems that require coordination across multiple teams and domains.
• Practical understanding of accelerator performance concepts (roofline analysis, memory bandwidth, occupancy) and what makes kernels perform on hardware.
• Track record of setting technical direction and mentoring engineers.
• Experience debugging performance issues that span kernel, runtime, compiler, or serving infrastructure.

• Hands-on experience with ML compiler stacks (torch.compile, TorchInductor, XLA, TVM, MLIR, Triton).
• Experience with autotuning, cost models, or search-based optimization (e.g., Ansor, AutoTVM, learned schedulers).
• Experience with production ML models (recommenders, LLMs, ranking) or hardware bring-up for accelerators.
• Kernel-level performance optimization experience: tiling, scheduling, memory layout, fusion.

Education Requirements

Bachelor's degree in Computer Science, Computer Engineering, Mathematics, or a related technical field, or equivalent practical experience. MS or PhD in CS, CE, compilers/systems, or related is preferred.

Compensation

United States: $154,003/year to $217,000/year (additional bonus, equity, and benefits)

About the Company

Company: Meta Platforms

Headquarters: Menlo Park, California, United States

American technology company that develops social networking products (Facebook, Instagram, WhatsApp) and invests in virtual/augmented reality hardware and software through Reality Labs, focusing on connectivity, advertising, and immersive computing experiences.