MWDN
MWDN
MWDN je podjetje za IT-storitve in razvoj programske opreme, ki podjetjem pomaga graditi in širiti digitalne izdelke za spletne, mobilne in namizne platforme. Njegova interna inženirska ekipa podpira izvedbo projektov za naročnike z razširjanjem ekip, namenskimi razvojnimi ekipami in globalnim zaposlovanjem IT-strokovnjakov. Z razvojnimi centri v srednji in vzhodni Evropi MWDN povezuje organizacije iz ZDA, Evrope in Izraela s strokovnjaki za programsko opremo za tehnološke projekte, prilagojene njihovim potrebam in stroškom.

CUDA Engineer, LLM Inference (Remote, Uzbekistan)

Optimize LLM inference on modern GPUs, working with CUDA kernels and serving frameworks. Join MWDN on AI systems and high-performance computing projects.

Opis

  • Improve LLM inference performance on modern GPUs.
  • Use inference and serving frameworks such as vLLM, SGLang, and TensorRT-LLM.
  • Examine GPU execution, memory access, kernel performance, and inference bottlenecks.
  • Build, tune, debug, or reverse-engineer CUDA kernels and low-level GPU operations.
  • Investigate optimization methods for matrix multiplication, attention, and neural network layers.
  • Benchmark systems and find ways to improve latency, throughput, and GPU utilization.
  • Tackle challenging problems spanning AI systems and high-performance computing.
  • Run practical experiments and examine low-level system behavior.

Zahteve

  • Bring substantial practical experience in CUDA and GPU programming.
  • Understand GPU architecture, memory hierarchies, parallel execution, and performance tuning in depth.
  • Have experience with LLM inference, AI systems, GPU kernels, or high-performance computing.
  • Know at least one of vLLM, SGLang, TensorRT-LLM, Triton, or FlashAttention.
  • Be able to diagnose complex performance problems and investigate low-level system behavior.
  • Apply strong problem-solving skills.
  • Communicate complex technical ideas effectively in English.
  • A PhD or research background in computer science, machine learning, computer architecture, or a related discipline is a plus.
  • Research publications, project work, or substantial open-source contributions to GPU or LLM infrastructure are a plus.
  • Experience with PyTorch internals, distributed inference, quantization, KV cache tuning, or custom GPU kernels is a plus.
  • Contributions to vLLM, SGLang, Triton, FlashAttention, or related repositories are a plus.

Ugodnosti

  • Client vetting designed to reduce risk and support reliable, timely payments.
  • Career guidance and help identifying future opportunities.
  • Legal advice on independent contracting or sole proprietorship, taxes, and related matters.
  • English lessons and opportunities for professional development.
  • Team-building activities.
  • Flexible working hours.
  • 29 days of paid time off: 18 working days per year plus national holidays.
  • 10 paid recovery days.
  • Financial and legal assistance for independent contractors.
  • Free English lessons with native English speakers or Ukrainian teachers.
  • Dedicated HR assistance.
  • People-first management with limited bureaucracy.
  • A welcoming company culture.

Sorodna delovna mesta

Ni najdenih delovnih mest

Poskusite prilagoditi ali počistiti filtre, da najdete več priložnosti.