Machine Learning Performance Engineer

Jane Street; New York (NY); On-site; USD 125,000 - 150,000; Full time

Job summary

An innovative firm is seeking a talented engineer to enhance their machine learning models' performance. This role involves optimizing training and inference processes, ensuring low-latency and high-throughput operations. As part of a dynamic ML team, you'll tackle complex challenges related to CUDA programming and system-level considerations. If you have a passion for solving intricate problems and a curious mindset, this could be the perfect opportunity for you to make a significant impact in the financial sector.

Qualifications

• Experience in low-level systems programming and optimisation.
• Understanding of modern ML techniques and toolsets.

Responsibilities

• Optimising the performance of ML models for training and inference.
• Improving CUDA performance and ensuring system efficiency.

Skills

Modern ML techniques

Low-level GPU knowledge

Debugging and optimisation

CUDA programming

Systems knowledge

Latency and throughput characteristics

Networking technologies

Collective algorithms

Tools

CUDA GDB

NSight Systems

NSight Compute

Triton

cuDNN

cuBLAS

Job description

We are looking for an engineer with experience in low-level systems programming and optimisation to join our growing ML team.

Machine learning is a critical pillar of Jane Street's global business. Our ever-evolving trading environment serves as a unique, rapid-feedback platform for ML experimentation, allowing us to incorporate new ideas with relatively little friction.

Your part here is optimising the performance of our models – both training and inference. We care about efficient large-scale training, low-latency inference in real-time systems and high-throughput inference in research. Part of this is improving straightforward CUDA, but the interesting part needs a whole-systems approach, including storage systems, networking and host- and GPU-level considerations. Zooming in, we also want to ensure our platform makes sense even at the lowest level – is all that throughput actually goodput? Does loading that vector from the L2 cache really take that long?

If you’ve never thought about a career in finance, you’re in good company. Many of us were in the same position before working here. If you have a curious mind and a passion for solving interesting problems, we have a feeling you’ll fit right in.

There’s no fixed set of skills, but here are some of the things we’re looking for:

• An understanding of modern ML techniques and toolsets
• The experience and systems knowledge required to debug a training run’s performance end to end
• Low-level GPU knowledge of PTX, SASS, warps, cooperative groups, Tensor Cores and the memory hierarchy
• Debugging and optimisation experience using tools like CUDA GDB, NSight Systems, NSight Compute
• Library knowledge of Triton, CUTLASS, CUB, Thrust, cuDNN and cuBLAS
• Intuition about the latency and throughput characteristics of CUDA graph launch, tensor core arithmetic, warp-level synchronization and asynchronous memory loads
• Background in Infiniband, RoCE, GPUDirect, PXN, rail optimisation and NVLink, and how to use these networking technologies to link up GPU clusters
• An understanding of the collective algorithms supporting distributed GPU training in NCCL or MPI
• An inventive approach and the willingness to ask hard questions about whether we're taking the right approaches and using the right tools