Meta Platforms · Menlo Park, CA

Software Engineer, Systems ML: PyTorch Distributed

Menlo Park, California, USFull-timeSoftware Engineering

Listed on Meta Platforms’s official careers site. Posted November 9, 2026. Check that it is still open and apply there.

Apply on Meta Platforms’s careers site →

Job description

In this role, you will be a member of the PyTorch Distributed team. The PyTorch team develops the open source software stack powering AI models and systems. The Distributed team builds the foundational infrastructure, APIs, kernels, tools, and recipes for enabling fast and reliable distributed training in PyTorch as well as in TorchTitan, the PyTorch-native framework for LLM pretraining and post-training. In this role, you will work closely with AI researchers & infrastructure engineers within Meta and from the open-source community to optimize distributed training using PyTorch.

A sample of the team’s recent work is highlighted in the following blogs: TorchTitan ([https://pytorch.org/blog/torchtitan-update-new-sota-models-performance-at-scale-titanrl/](https://pytorch.org/blog/torchtitan-update-new-sota-models-performance-at-scale-titanrl/)), DistMoE ([https://meta-pytorch.org/dist_moe/previews/visual-explainer/](https://meta-pytorch.org/dist_moe/previews/visual-explainer/)), OLMO3 reproduction ([https://docs.pytorch.org/devlogs/distributed/2026-09-02-olmo3-reproduction/](https://docs.pytorch.org/devlogs/distributed/2026-09-02-olmo3-reproduction/)).

Responsibilities

  • Improve PyTorch's state of the art training, post-training, and inference on modern AI hardware accelerators Enable new models in TorchTitan for pre-training and post-training
  • Optimize the performance of TorchTitan’s pre-training and post-training of latest Gen AI models
  • Develop the latest parallelism APIs, kernels, and tools for debugging/visualizing large distributed training jobs
  • Build foundational distributed training infrastructure including communication collectives, profiling, memory allocation/management, etc

Qualifications

Currently has, or is in the process of obtaining a Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience. Degree must be completed prior to joining Meta Proven Python programming Experience in AI framework development or accelerating deep learning models on hardware architectures Experience adhering to and implementing responsible, ethical AI practices (e.g., risk assessment, bias mitigation, quality and accuracy reviews)

Knowledge of GPU, CPU, or AI hardware accelerator architectures Experience working with frameworks like PyTorch, Caffe2, TensorFlow, ONNX, TensorRT Demonstrated ongoing AI skill development (e.g., prompt/context engineering, agent orchestration) and staying current with emerging AI technologies OR AI high performance kernels: Experience with CUDA programming, OpenMP / OpenCL programming or AI hardware accelerator kernel programming.

Experience in accelerating libraries on AI hardware, similar to cuBLAS, cuDNN, CUTLASS, HIP, ROCm etc OR AI frameworks: Experience in developing training and inference framework components. Experience in system performance optimizations such as runtime analysis of latency, memory bandwidth, I/O access, compute utilization analysis and associated tooling development Demonstrated ability to integrate AI tools to optimize/redesign workflows and drive measurable impact (e.g., efficiency gains, quality improvements)

More jobs at Meta Platforms

Software Engineer - Android

Menlo Park, California, US

Software Engineer, Product

Menlo Park, California, US

Research Scientist, PAR Safety

Menlo Park, California, US

Similar jobs at other companies

Principal Software Engineer - PyTorch Training Frameworks

Advanced Micro Devices (AMD) · San Jose, California

Software and Data Engineer - Software Engineer III- Agentic Pricing

JPMorganChase · Jersey City, NJ, United States

OpenNash lists jobs from employers’ own careers sites and does not handle applications. Dates come from the employer where stated; “first seen” and “last checked” are our own observations.