Machine learning engineer for LLMs

This isn't another ML role where you're calling other models, you'll be actively training the language models that millions of people use when they shop online. This AI team does more than call APIs. It does continued pre-training on 100B+ tokens of its own data, fine-tunes models, and puts them into production on its own GPU cluster.

• Train LLMs with continued pre-training and full fine-tuning on large proprietary datasets
• Build and optimise distributed training on multi-node H200 clusters (DeepSpeed, FSDP, Megatron-LM)
• Own data at scale: filtering, deduplication, quality and data mixtures for corpora of 100B+ tokens
• Generate synthetic data for instruction tuning, and build the evaluation frameworks that guide decisions
• Lead the technical work: make architecture decisions, review code and mentor ML engineers

• 7+ years in ML engineering. You have trained language models and put them into production yourself (LLMs, encoders or embedding models)
• Hands-on experience with multi-node training. You know how to fix loss spikes, NCCL timeouts and slow throughput
• Expert in Python and PyTorch (Transformers, Accelerate, DeepSpeed)
• Experience as a technical lead. You make others better and focus on results for the business
• Будет плюсом: DPO/GRPO/RLHF, vLLM and distillation, papers at NeurIPS/ICML/ACL/EMNLP, or models and PRs on Hugging Face

• €90k–185k base + 20% bonus