About Nebius
Nebius is leading a new era in cloud infrastructure for the global AI economy, building a full-stack AI cloud platform. Listed on Nasdaq (NBIS) and headquartered in Amsterdam.
The Role
Token Factory is part of Nebius Cloud, one of the world's largest GPU clouds, running tens of thousands of GPUs, building an inference and fine-tuning platform that makes every kind of foundation model fast, reliable, and effortless to train and deploy at massive scale.
Current Directions
- Advanced Fine-Tuning: enhancing LoRA-based and full-parameter fine-tuning methodologies for cutting-edge LLMs, focusing on model quality and training efficiency.
- Inference Optimization: identifying LLM inference bottlenecks, building training/evaluation pipelines in JAX for speculative decoding, and deriving scaling laws.
- Low Precision Training & Inference: investigating FP8/NVFP4/MXFP4 methodologies for supervised fine-tuning and reinforcement learning, optimized for modern hardware.
We Expect You to Have
- A profound understanding of theoretical foundations of machine learning and reinforcement learning.
- Deep expertise in modern deep learning for language processing and generation.
- Experience training large models on multiple computational nodes, with understanding of sharding strategies and custom kernels.
- Strong software engineering skills (mostly Python) and deep experience with JAX.
Nice to Have
Experience with language models or NLP, familiarity with MHA, RoPE, ZeRO/FSDP, Flash Attention, and quantization, plus a track record building products in a startup-like environment.