Description
We are looking for a senior ML compiler developer (C++/MLIR/OpenXLA/IREE) — a person who can transform ML model graphs into fast code for real accelerators, is not afraid to dig into low-level optimization and IR levels. You will work at the intersection of hardware and software and participate in the co-design of the entire architecture.
Responsibilities
- development of an ML compiler based on MLIR: custom MLIR passes, IR, lowering, codegen
- optimization for photonic accelerators: fusion, scheduling, cost model, memory planning
- performance profiling on real ML models, taking into account the specifics of the photonic accelerator
- participation in the design and development of system-level software for the photonic accelerator
- integration of the compiler toolchain with PyTorch.
Requirements
- excellent knowledge of C++, confident knowledge of Python
- practical experience with MLIR/OpenXLA/IREE/TVM/OpenVINO (not necessarily all), ability to understand performance at a low level.
- understanding of accelerator architectures (GPU/TPU/DSA).
Desirable:
- experience in optimizing ML for GPU/TPU/other accelerators
- practical experience in designing ISA/opcodes/memory/instructions.
Conditions
- hybrid format, 3+ days in the office (Tulskaya metro station)
- strong team (ex-Samsung, ex-Yandex)
- lab and hardware at the next table - we push changes to production quickly
- annual salary review and annual bonus
- corporate gym and recreation areas
- more than 400 educational programs from SberUniversity for professional and career development
- extended voluntary health insurance, preferential insurance for family and corporate pension program
- flexible mortgage discount equal to 1/3 of the Central Bank's key rate
- free SberPrime+ subscription, discounts on products from partner companies.