In this role, you will be a member of the Network.AI Software team and part of the bigger DC networking organization. The team develops and owns the software stack around NCCL (NVIDIA Collective Communications Library), which enables multi-GPU and multi-node data communication through HPC-style collectives. NCCL has been integrated into PyTorch and is on the critical path of multi-GPU distributed training. In other words, nearly every distributed GPU-based ML workload in Meta Production goes through the SW stack the team owns.At the high level, the team aims to enable Meta-wide ML products and innovations to leverage our large-scale GPU training and inference fleet through an observable, reliable and high-performance distributed AI/GPU communication stack. Currently, one of the team’s focus is on building customized features, SW benchmarks, performance tuners and SW stacks around NCCL and PyTorch to improve the full-stack distributed ML reliability and performance (e.g. Large-Scale GenAI/LLM training) from the trainer down to the inter-GPU and network communication layer. And we are seeking for engineers to work on the space of GenAI/LLM scaling reliability and performance.

Software Engineer, SystemML - Scaling / Performance Responsibilities:

• Enabling reliable and highly scalable distributed ML training on Meta's large-scale GPU training infra with a focus on GenAI/LLM scaling

Minimum Qualifications:

• Specialized experience in one or more of the following machine learning/deep learning domains: Distributed ML Training, GPU architecture, ML systems, AI infrastructure, high performance computing, performance optimizations, or Machine Learning frameworks (e.g. PyTorch).

Preferred Qualifications:

• PhD in Computer Science, Computer Engineering, or relevant technical field
• Experience with NCCL and distributed GPU reliability/performance improvment on RoCE/Infiniband
• Experience working with DL frameworks like PyTorch, Caffe2 or TensorFlow
• Experience with both data parallel and model parallel training, such as Distributed Data Parallel, Fully Sharded Data Parallel (FSDP), Tensor Parallel, and Pipeline Parallel
• Experience in AI framework and trainer development on accelerating large-scale distributed deep learning models
• Experience in HPC and parallel computing
• Knowledge of GPU architectures and CUDA programming
• Knowledge of ML, deep learning and LLM

About Meta:
Meta builds technologies that help people connect, find communities, and grow businesses. When Facebook launched in 2004, it changed the way people connect. Apps like Messenger, Instagram and WhatsApp further empowered billions around the world. Now, Meta is moving beyond 2D screens toward immersive experiences like augmented and virtual reality to help build the next evolution in social technology. People who choose to build their careers by building with us at Meta help shape a future that will take us beyond what digital connection makes possible today—beyond the constraints of screens, the limits of distance, and even the rules of physics.