Software Engineer (ML Training Performance)
Posted
Description
- Are you excited to drive our ML Training Optimization initiatives and make the models that power autonomous driving trains faster, cheaper, and at greater scale? You will work with SOTA accelerators and cutting-edge techniques in distributed training, parallelism strategies, mixed-precision training and RL, especially for VLA and VLM models that are multi-modal — partnering closely with all the Autonomy teams within Zoox (Perception, Prediction, Planner, Simulation, Collision Avoidance) and pushing the boundaries of how ML training is practiced at Zoox
- We build and operate the base layer of ML tools, model development, and training systems that our applied research teams use for in- and off-vehicle ML use cases. You will work alongside a team of strong software engineers and act as a force multiplier for our internal customers. This team has many growth opportunities as we expand our robotaxi deployments and venture into new ML domains. If you want to learn more about our stack behind autonomous driving, please look here. If you want to learn more about our ML Infrastructure, here is one of our past talks at re:Invent
- Design, implement, and operate cutting-edge ML training performance optimization techniques — distributed training (data/tensor/pipeline parallelism), mixed-precision, gradient checkpointing, and GPU kernels — to scale our VLM, VLA, and Foundational models training efficiency
- Work on RL framework and RL performance optimization for driving foundation model
- Profile and debug training throughput bottlenecks across the compute, compiler, framework, and cluster-networking stack
- Collaborate closely with ML researchers, data engineers, and hardware engineers to define requirements and align on architectural decisions for large-scale training infrastructure
- Implement custom kernels and build benchmarks to track and validate training-performance optimizations across the stack
Benefits
- Paid parental leave
- Affinity groups and sports clubs
- Work from home opportunities
- Health insurance
- Our crew’s health and happiness is our first priority. We offer comprehensive health and mental health support, a wellbeing program, and unlimited and flexible paid time away
- We invest in our crew—and their families—for the long term. That includes generous family planning support, caregiver support, and strong cash compensation with great equity upside
- We look after our crew when they’re in the office too. Our famous food program is a great example, featuring a daily changing menu of local and sustainable dishes
- There’s a busy calendar of social events at Zoox, with more sports teams than you can count. And, of course, playing with robots is an important part of the job description
- Experience with RL framework such as Verl/MILES and RL algorithms(GRPO, PPO)
- Proficient in Python or C++
- Experience with collective communication libraries (e.g., NCCL) and optimizing communication/compute overlap in multi-node training
- Experience with distributed training frameworks (Megatron-LM, TorchTitan), efficiently utilizing GPU/accelerator clusters at scale
- 4+ years of total experience, including 2+ years working on large-scale model training platforms
- Experience using profiling tools (NVIDIA Nsight, PyTorch Profiler, or XLA/JAX profiling) to identify and resolve training bottlenecks
- Experience with ML compilers, performance tooling, or hardware/software co-design (e.g., GPU/TPU/accelerator architectures)
- You do not need to match every listed expectation to apply for this position
Skills
Sign in to read and join the discussion.