ML Engineer

The Token Factory team is part of Nebius Cloud, a leading GPU cloud platform. We are building a high-performance inference and fine-tuning platform to push foundation models to their hardware limits, with the goal of maximum throughput, minimal latency, and cost-efficiency across thousands of GPUs. The role involves work in inference optimization, inference engine support, and low-precision training/inference pipelines. You will identify bottlenecks in LLM inference to speed up production, implement speculative decoding architectures, and contribute to open-source inference engines. You will design and productionise low-precision training/inference pipelines with measurable throughput/cost gains. Requirements include deep understanding of ML theory and transformer architectures, experience profiling GPU workloads with Nsight or PyTorch profiler, knowledge of GPU memory hierarchy, familiarity with MHA, RoPE, KV-cache, Flash Attention, and quantisation, and strong Python software engineering skills. Nice-to-have items include experience with open-source inference engines, kernel languages or DSLs like Triton, CUDA, a track record in dynamic startup-like environments, and contributions to open-source projects. Benefits include competitive compensation, career growth, flexibility, collaborative culture, and international teams. Nebius emphasizes equal opportunity and inclusive hiring. Remote Europe is available for this role.