Senior Deep Learning Compiler Engineer - PyTorch

Role Overview

Join NVIDIA's pioneering team at the cutting edge of AI compiler technology, focused on building groundbreaking tools for accelerating computing. You will participate in developing Thunder, an innovative source-to-source compiler designed to maximize performance of PyTorch models on NVIDIA GPUs. This position offers the chance to influence the PyTorch ecosystem substantially by collaborating on modern compiler stacks, including PyTorch 2.0's TorchDynamo and TorchInductor, creating impactful open-source solutions for global AI developers and researchers.

Primary Responsibilities

In this technically demanding role, you will spearhead the design, development, and refinement of key compiler technologies that enable accelerated deep learning workloads. Collaboration with NVIDIA’s PyTorch engineering team is central, contributing to framework advancements and optimization strategies. Your efforts will include deep performance evaluation across extensive GPU workloads to identify and implement enhancements shaping the evolution of Thunder. Partnership with teams specializing in compilers, libraries, and systems such as nvFuser, TVM, XLA, and CUDA is integral, turning cutting-edge research into tangible benefits for the community.

Candidate Profile

• Educational background includes Bachelor’s, Master’s, or Ph.D. in Computer Science or related fields, or equivalent experiential qualifications.
• 8 or more years of substantive professional experience.
• Proficient in Python with a history of constructing robust, thoroughly tested software architectures.
• Practical experience working with deep learning frameworks like PyTorch or JAX, with insights into model creation and identifying performance bottlenecks.
• Strong understanding of compiler theory, covering abstract syntax trees (AST), intermediate representations like SSA, program analysis methods, and code generation techniques.
• Exceptional communication and teamwork capabilities, vital for successful collaboration in distributed and open-source settings.

Preferred Qualifications

• Prior contributions to deep learning compiler projects (e.g., TVM, MLIR, IREE) or core deep learning frameworks.
• In-depth knowledge of PyTorch internals, especially its compiler components such as TorchDynamo and TorchInductor.
• Familiarity with JAX-style functional transformations applied within compiler environments.
• Experience with parallel and distributed programming, including development of optimized CUDA code.
• Active involvement in open-source communities through code contributions, design input, or mentorship roles.

Company Commitment and Compensation

NVIDIA is a leader in AI, High-Performance Computing, and Visualization, employing innovative and driven professionals dedicated to technological advancement. The company values diversity, inclusion, and offers a supportive workplace with competitive compensation and extensive benefits. This role’s salary will be calculated based on experience, location, and comparable position salaries, with the specified range in Poland from 292500 to 507000 PLN annually.