ML Systems Engineer - Accelerate GPU Training (Equity)

Digital Waffle; Menlo Park (CA); On-site; USD 350,000 - 550,000; Full time

Benefits offered by this job

Equity

Job summary

Digital Waffle is seeking a Member of Technical Staff, ML Systems in San Francisco (On-site). You will optimize GPU and system performance for image, video and world-model workloads, and build distributed inference and training engines across multiple GPUs.

You will profile bottlenecks with Nsight, write low-level CUDA and Triton code, and own communication performance (NCCL, RDMA over InfiniBand/RoCE) to accelerate production workloads, reporting directly to the CEO.

Qualifications

• 1+ years in deep learning training or inference systems, or distributed systems.
• Hands-on experience with CUDA, Triton, PyTorch, NCCL or RDMA.
• Excited about shaving milliseconds off production workloads.

Responsibilities

• Optimize GPU and system performance for image, video and world-model training and inference.
• Profile and remove bottlenecks at kernel, memory, system and cluster level using Nsight and related tooling.
• Write low-level CUDA and Triton optimizations on code paths that run in production.
• Build distributed inference and training engines for diffusion models across multiple GPUs and nodes.
• Own communication performance: NCCL, RDMA over InfiniBand/RoCE, and disaggregated serving.
• Build benchmarking and regression harnesses so performance gains stick.

Skills

Deep learning systems

Distributed systems

Performance optimization

Tools

CUDA

Triton

PyTorch

NCCL

RDMA

Job description

Digital Waffle is seeking a Member of Technical Staff, ML Systems in San Francisco (On-site). You will optimize GPU and system performance for image, video and world-model workloads, and build distributed inference and training engines across multiple GPUs.

You will profile bottlenecks with Nsight, write low-level CUDA and Triton code, and own communication performance (NCCL, RDMA over InfiniBand/RoCE) to accelerate production workloads, reporting directly to the CEO.