Research Scientist Intern, AI & Compute Foundation - MTIA SW (PhD)

Meta•Menlo Park, CA

1d•$7,650 - $12,667

About The Position

The MTIA (Meta Training & Inference Accelerator) Software team is part of the AI & Compute Foundation organization. It explores, develops, and helps productize high-performance software and hardware technologies for AI at datacenter scale. The team co-optimizes software, including algorithms and numerics, and hardware, including platforms and networks, to create balanced system designs. It also studies performance bottlenecks on existing systems and optimizes AI production models, contributing to total cost of ownership improvements for key AI services. The team develops AI frameworks to accelerate Meta’s deep learning and machine learning workloads on MTIA accelerator hardware. Its software stack includes kernel libraries that use hardware architectural features to support high-performance inference and training workloads. The internship lasts 12 to 24 weeks, with various start dates throughout the year. Interns develop software in areas such as AI frameworks, compiler stacks, high-performance kernels, and acceleration for next-generation hardware; contribute to PyTorch core compilers; analyze deep learning networks and implement compiler optimizations; collaborate with AI research scientists; and tune deep learning frameworks and software components.

Requirements

Be in the process of obtaining a PhD in Computer Science or a related STEM field.
Have research experience in one or more of the following: AI compilers, high-performance kernel development, deep learning framework and runtime internals, numerics and low-precision formats, or parallelism strategies for large-scale training and inference.
Have knowledge of machine learning systems and computer architecture, including memory and storage hierarchies, parallel execution, and performance analysis on accelerated hardware.
Demonstrate programming skills in C/C++, Python, or Rust.
Obtain work authorization in the country of employment by the time of hire and maintain authorization throughout employment.

Nice To Haves

For AI compiler work: experience with compiler optimizations such as loop optimization, vectorization, parallelization, or hardware-specific optimization such as SIMD; experience with MLIR, LLVM, IREE, XLA, TVM, or Halide is a plus.
For AI framework work: experience developing training and inference framework components such as SGLang, vLLM, PyTorch, Caffe2, TensorFlow, ONNX, or TensorRT.
For high-performance AI kernels: experience with CUDA, OpenMP, OpenCL, or AI accelerator kernel programming; experience accelerating libraries on AI hardware, such as cuBLAS, cuDNN, CUTLASS, HIP, or ROCm.
Experience with system performance optimization, including runtime analysis of latency, memory bandwidth, I/O access, and compute utilization, and related tooling development.
Knowledge of GPU, CPU, or AI accelerator architectures.
A record of results demonstrated by publications at leading systems conferences such as MLSys, OSDI, ASPLOS, ISCA, or CGO, or systems-focused work at ML venues such as NeurIPS, ICML, or ICLR.
Experience working and communicating cross-functionally in a team environment.
Intent to return to a degree program after completing the internship or co-op.

Responsibilities

Develop software stack components focused on AI frameworks, compiler stacks, high-performance kernel development, or acceleration for next-generation hardware architectures.
Contribute to PyTorch core compilers to support state-of-the-art inference and training accelerators and optimize their performance.
Analyze deep learning networks and develop and implement compiler optimization algorithms.
Collaborate with AI research scientists to accelerate deep learning models, including recommendation systems, generative AI, computer vision, and NLP.
Tune and optimize deep learning frameworks and software components.