ML Performance Engineer — Hybrid, High-Impact Role

Trading Interview; New York, Northern (NY, KY); Hybrid; USD 180,000 - 220,000; Full time

Benefits offered by this job

Hybrid working opportunities

Generous time off

Free meals daily

Wellness reimbursement

Volunteer opportunities

Job summary

Tower Research Capital in New York seeks a Training Performance and Distributed Training Engineer to accelerate ML model training at scale. You will optimize end-to-end training pipelines from data ingestion to kernel performance, enabling researchers to iterate on increasingly complex models.

You will benchmark across CPUs/GPUs, design distribution strategies, and collaborate with ML researchers, HPC engineers, and infrastructure teams to improve throughput, stability, and cost efficiency.

Qualifications

• 3+ years optimizing ML training workloads in distributed or large-scale computing environments.
• Deep knowledge of PyTorch or JAX, including distributed-training capabilities.
• Strong Python and C++ programming for performance-critical systems.

Responsibilities

• Benchmark training workloads across CPUs, GPUs, and accelerators to identify bottlenecks.
• Design and optimize distributed training strategies (data, tensor, pipeline, model parallelism).
• Improve end-to-end training efficiency (data loading, memory management, checkpointing).
• Collaborate with ML Researchers, HPC Engineers, and infrastructure teams to improve throughput and reliability.

Skills

Python

C++

Distributed training

Performance optimization

Education

Bachelor's degree in CS/EE/Math

Tools

PyTorch

JAX

CUDA

Triton

CUTLASS

cuBLAS

cuDNN

NCCL

DeepSpeed

Job description

Tower Research Capital in New York seeks a Training Performance and Distributed Training Engineer to accelerate ML model training at scale. You will optimize end-to-end training pipelines from data ingestion to kernel performance, enabling researchers to iterate on increasingly complex models.

You will benchmark across CPUs/GPUs, design distribution strategies, and collaborate with ML researchers, HPC engineers, and infrastructure teams to improve throughput, stability, and cost efficiency.