California jobs

Member of Technical Staff, Distributed Training Performance

HarkSan Jose +1 placeFull-timeSan Jose · San Jose, CAPosted 11 days ago

$180k–$450k/yr listed by the employer

Looks real. 1 thing to check before you apply.

On Hark’s own careers site (Greenhouse), checked open today.

  • To check: Wide pay range: top of the range is 2.5× the bottomAsk what the job pays at your level before you interview.
  • Checked: On the company's own careers siteGreenhouse board
  • Checked: Checked open todayGreenhouse board
  • Checked: Pay listedJob posting
  • Checked: No reposts foundGreenhouse board
  • Trust score 100 / 100 · Strong
See all 6 checksShow fewer
How we check

What you’ll do

  • Own and improve MFU and step time for large-scale dense and mixture-of-experts pretraining runs.
  • Develop and optimize GPU kernels, parallelism strategies, communication overlap, memory use, and activation recomputation.
  • Profile training workloads and diagnose performance regressions, numerical instability, failures, and hardware or node issues.
  • Build reusable training components for new architectures and multimodal speech, vision, and text encoders.

Summarised from the posting by AI and can be wrong. The full description is the source.

Requirements

Experience

  • Experience developing distributed training systems and working with the internals of Megatron-LM, NeMo, or a comparable framework, including its parallelism strategies
  • Experience using GPU profilers to analyze kernel execution, memory bandwidth, communication overhead, and data-loading stalls
  • Experience developing CUDA or Triton kernels, or using CUTLASS, for attention, normalization, or mixture-of-experts models
  • A record of measurably improving MFU or throughput on large-scale training jobs
  • Experience with mixture-of-experts training at scale, including expert parallelism, load balancing, and all-to-all communication optimization (preferred)
  • Experience with GPU communication libraries such as NCCL or network topology-aware scheduling (preferred)
  • Experience with fault-tolerant training, asynchronous checkpointing, in-memory recovery, or elastic training (preferred)
  • Experience balancing workloads across different encoders in multimodal training (preferred)
  • Familiarity with cluster schedulers (Slurm, Kubernetes) and GPU health monitoring at fleet scale (preferred)

Skills

  • Working knowledge of mixed-precision training, including BF16 and FP8, and methods for diagnosing numerical instability

Your match

Sign in and add your resume to see which of this job's requirements it shows.

Sign in to check your match

Saved to your account, encrypted. The AI that reads it doesn't keep it or train on it.

Full job description

From Hark’s careers site

About Hark

Hark is an artificial intelligence company building advanced, personalized intelligence. One that is proactive, multimodal, and capable of interacting with the world through speech, text, vision, and persistent memory.

We're pairing that intelligence with next-generation hardware to create a universal interface between humans and machines. While today's AI largely operates through chat boxes and decade-old devices, Hark is focused on what comes next: agentic systems that interact naturally with people and the real world.

Show full descriptionShow less

To get there, we're developing multimodal models and next-generation AI hardware together - designed from the ground up as a single, unified interface for a new era of intelligent systems.

About the Role

The Omni team at Hark is building the next generation of AI experiences beyond text, enabling models to understand and generate content across multiple modalities, including text, audio, and beyond. Our goal is to create seamless, real-time multimodal intelligence that powers intuitive and immersive user experiences.

As part of the Omni team, you'll improve the performance of Hark's Megatron-based pretraining systems across thousands of GPUs. The role focuses on model FLOPs utilization (MFU), training throughput, and the performance of kernels, communication, and parallelism strategies. You'll work with researchers to support new architectures and with infrastructure engineers to resolve bottlenecks, maintain numerical stability, and keep large runs reliable.

Responsibilities

  • Own MFU and step time for pretraining runs across dense and mixture-of-experts models, setting targets per model size and tracking them across runs.
  • Write or integrate fused and custom kernels (CUDA, Triton, CUTLASS) for attention, normalization, and mixture-of-experts routing.
  • Overlap communication with compute, reduce pipeline bubbles, and tune activation recomputation and memory use to raise throughput per GPU.
  • Implement model and optimizer sharding, and optimize data, tensor, pipeline, sequence, context, and expert parallelism for dense and mixture-of-experts architectures.
  • Profile GPU kernels, collective communication, data loading, and mixed-precision training to identify performance and numerical stability bottlenecks.
  • Develop reusable training components that support new architectures and multimodal encoders for speech, vision, and text.
  • Investigate performance regressions and training failures involving communication, numerical changes, hardware faults, or straggling nodes.
  • Work with architecture researchers and compute infrastructure engineers to align model designs, training software, and cluster capabilities.
  • Improve checkpointing, restart time, failure detection, and observability to reduce lost training work. Participate in on-call support during large runs.

Requirements

  • Experience developing distributed training systems and working with the internals of Megatron-LM, NeMo, or a comparable framework, including its parallelism strategies.
  • Strong understanding of data, tensor, pipeline, sequence, context, and expert parallelism, including communication patterns, memory tradeoffs, and failure modes.
  • Experience using GPU profilers to analyze kernel execution, memory bandwidth, communication overhead, and data-loading stalls.
  • Working knowledge of mixed-precision training, including BF16 and FP8, and methods for diagnosing numerical instability.
  • Strong Python and C++/CUDA engineering skills, with experience building reliable tools for researchers.
  • Experience developing CUDA or Triton kernels, or using CUTLASS, for attention, normalization, or mixture-of-experts models.
  • A record of measurably improving MFU or throughput on large-scale training jobs.

Bonus Qualifications

  • Experience with mixture-of-experts training at scale, including expert parallelism, load balancing, and all-to-all communication optimization.
  • Experience with GPU communication libraries such as NCCL or network topology-aware scheduling.
  • Experience with fault-tolerant training, asynchronous checkpointing, in-memory recovery, or elastic training.
  • Experience balancing workloads across different encoders in multimodal training.
  • Familiarity with cluster schedulers (Slurm, Kubernetes) and GPU health monitoring at fleet scale.

Compensation

The US base salary range for this full-time position is between $180,000 - $450,000 annually.

The pay offered for this position may vary based on several individual factors, including job-related knowledge, skills, and experience. The total compensation package may also include additional components/benefits depending on the specific role. This information will be shared if an employment offer is extended.

Something wrong with this job? Report it

Similar jobs

See more like this