обновлено 16 дней назад
NPU SDK Software Engineer (AI)
Вакансия из Hirify Global, списка международных tech-компанийДля мэтча и отклика нужен Plus
Мэтч & Сопровод
Для мэтча с этой вакансией нужен Plus
Описание вакансии
Текст:
TL;DR
NPU SDK Software Engineer (AI): Own end-to-end SDK user experience for bringing state-of-the-art AI models (LLM, Multi-modal, Vision, Speech) onto Rebellions NPU with an accent on model porting, CI/CD pipelines, framework integration, and developer documentation. Focus on optimizing performance, ensuring numerical parity, designing verification tools, and integrating with Hugging Face, vLLM ecosystems.
Location: office in Seongnam-si, Bundang-gu, South Korea (경기도 성남시 분당구 정자일로156번길 6, R-TOWER 3F ~ 8F)
Company
AI hardware company developing Neural Processing Units (NPU) for edge AI inference.
What you will do
- Design, build, and operate CI/CD pipelines to automate SDK release processes
- Port and optimize deep learning models (LLM, Multi-modal) onto NPU, analyze bottlenecks, validate against GPU references
- Integrate SDK with open-source ecosystems like Hugging Face Transformers, Diffusers, vLLM
- Design model-feeding frameworks and debugging utilities for functional correctness and stability
- Author developer documentation including API references, tutorials, model matrices, release notes
Requirements
- Master’s degree or above in Computer Science, Electrical Engineering, or related field
- Deep knowledge of CI/CD tools (GitHub Actions, Airflow, Buildkite)
- Strong understanding of deep learning architectures (LLMs, Multi-modal, Vision, Speech)
- Familiarity with PyTorch internals, model customization, graph transformations
- High-performance development in Python and Modern C++ (C++17+)
- Experience with Hugging Face libraries and vLLM integration
- Hands-on with Python and Kubernetes for AI inference workloads
Nice to have
- 5+ years relevant experience
- Model optimization on hardware (CUDA, TensorRT, MLIR, Triton, TVM)
- Scalable ML infrastructure with Docker/Kubernetes
- AI Harness Engineering experience
- Open-source contributions
- Knowledge of JAX or TensorFlow internals
Hiring process
- Document screening > Online interview > On-site interview > Culture-fit interview > Compensation discussion > Final offer
- Process may vary by role and schedule
- Results sent via email
Будьте осторожны: если работодатель просит войти в их систему, используя iCloud/Google, прислать код/пароль, запустить код/ПО, не делайте этого - это мошенники. Обязательно жмите "Пожаловаться" или пишите в поддержку. Подробнее в гайде →